Downstream.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "9ef08229-5db6-424e-8cc6-54bf176a8dc0",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "import math\n",
    "import itertools\n",
    "import collections\n",
    "import re\n",
    "import os\n",
    "\n",
    "import tqdm\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import scipy\n",
    "\n",
    "import statsmodels\n",
    "import statsmodels.api as sm\n",
    "import statsmodels.formula.api as smf\n",
    "\n",
    "import sklearn\n",
    "from sklearn import ensemble\n",
    "from sklearn import pipeline\n",
    "from sklearn.metrics import accuracy_score, roc_curve, roc_auc_score, precision_recall_curve, average_precision_score, f1_score\n",
    "from sklearn.model_selection import cross_val_score, cross_validate, cross_val_predict\n",
    "from sklearn.cross_decomposition import PLSRegression\n",
    "from sklearn.metrics.pairwise import cosine_similarity\n",
    "import patsy\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib as mpl\n",
    "from matplotlib import cm\n",
    "from matplotlib.colors import ListedColormap, LinearSegmentedColormap\n",
    "import seaborn as sns\n",
    "#import statannot\n",
    "import statannotations\n",
    "from statannotations.Annotator import Annotator\n",
    "from statannotations.stats.StatTest import StatTest\n",
    "import plotnine as p9\n",
    "\n",
    "import Bio.motifs as motifs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "9a77ab9b-1af3-41f7-8b6a-1c568c3e09d7",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "# paths \n",
    "project_path = \"./outputs/\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8653a513-c32b-4355-8acb-ccbdb3fd7738",
   "metadata": {},
   "source": [
    "# Utility Functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "c5ce82f2-6fd9-4deb-b0a6-f850efafe743",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "class Kmerizer:\n",
    "    \n",
    "    def __init__(self, k, overlap=True, log=False, divide=False, leave_one_out=False):\n",
    "        self.k = k\n",
    "        if leave_one_out:\n",
    "            self.kmers = {\"\".join(x):i for i,x in zip(range(4**k), [x for x in itertools.product(\"ACGT\",repeat=k)][:-1])}\n",
    "        else:\n",
    "            self.kmers = {\"\".join(x):i for i,x in zip(range(4**k), itertools.product(\"ACGT\",repeat=k))}\n",
    "        self.log = log\n",
    "        self.divide = divide\n",
    "        self.overlap = overlap\n",
    "        \n",
    "    def kmerize(self, seq):\n",
    "        counts = np.zeros(4**self.k)\n",
    "        i = 0\n",
    "        jump = 1 if self.overlap else self.k\n",
    "        while i < len(seq) - self.k + 1: \n",
    "            kmer = seq[i:i+self.k]\n",
    "            if \"N\" in kmer:\n",
    "                i += jump\n",
    "                continue\n",
    "            counts[self.kmers[kmer]] += 1\n",
    "            i += jump\n",
    "        if self.divide:\n",
    "            counts = counts/len(seq)\n",
    "        if self.log:\n",
    "            counts = np.log(counts + 1)\n",
    "        return counts\n",
    "    \n",
    "    def tokenize(self, seq, jump=False):\n",
    "        kmers = []\n",
    "        i = 0\n",
    "        while i < len(seq) - self.k + 1: \n",
    "            kmer = seq[i:i+self.k]\n",
    "            kmers.append(kmer)\n",
    "            if jump:\n",
    "                i += self.k\n",
    "            else:\n",
    "                i += 1\n",
    "        return kmers\n",
    "    \n",
    "kmerizer2 = Kmerizer(k=2)\n",
    "kmerizer3 = Kmerizer(k=3)\n",
    "kmerizer4 = Kmerizer(k=4)\n",
    "kmerizer5 = Kmerizer(k=5)\n",
    "kmerizer6 = Kmerizer(k=6)\n",
    "kmerizer7 = Kmerizer(k=7)\n",
    "\n",
    "codonmerizer = Kmerizer(k=3,overlap=False, log=True, divide=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "4a9f9e2c-9054-491d-ab03-a1dc49d8ce56",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def pearson_scorer(estimator, X, y):\n",
    "    y_pred = estimator.predict(X)\n",
    "    if len(y_pred.shape) == 2:\n",
    "        y_pred = y_pred.reshape(-1)\n",
    "    return scipy.stats.pearsonr(y, y_pred)[0]\n",
    "\n",
    "def pearson_r2(estimator, X, y):\n",
    "    y_pred = estimator.predict(X)\n",
    "    #print(estimator[1].alpha_)\n",
    "    if len(y_pred.shape) == 2:\n",
    "        y_pred = y_pred.reshape(-1)\n",
    "    return scipy.stats.pearsonr(y, y_pred)[0] ** 2\n",
    "\n",
    "def pearson_r2_metric(y_true, y_pred):\n",
    "    return scipy.stats.pearsonr(y_true, y_pred)[0] ** 2\n",
    "\n",
    "def spearman_metric(y_true, y_pred):\n",
    "    return scipy.stats.spearmanr(y_true, y_pred)[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "ca1d9a75-0692-42ab-a2c2-9101022983f3",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "import os\n",
    "def extract_model_and_dataset_from_directory(dir_path):\n",
    "    dataset_and_model_map = dict()\n",
    "    for folder in os.listdir(dir_path):\n",
    "        try:\n",
    "            _, _, model_type, dataset, _ = folder.split('-')\n",
    "        except:\n",
    "            continue\n",
    "        if not dataset in dataset_and_model_map.keys():\n",
    "            dataset_and_model_map[dataset] = [folder]\n",
    "        else:\n",
    "            dataset_and_model_map[dataset].append(folder)\n",
    "    return dataset_and_model_map"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "bef7f77e-2065-4103-b0e9-33662f1de581",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "dataset_and_model_map = extract_model_and_dataset_from_directory('./outputs/')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "d6ccf746-3073-4d22-a811-ef615e93bb70",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def get_model_name(path):\n",
    "    token = \" \".join(path.split(\"-\")[2].split(\"_\")[1:])\n",
    "    model_name  = path.split(\"-\")[1]\n",
    "    yeast = \"yeast only, \" if \"yeast\" in path else \"\" \n",
    "    sacc = \"sacc only, \" if \"_sacc_incl\" in path else \"\"\n",
    "    token = f\"Aware:{token}, \" if not 'agnostic' in token else f\"Agnostic, \"\n",
    "    sacc_incl = \"including all saccs\" if \"incl_sacc\" in path else \"\"\n",
    "    sacc_incl = \"\" if \"sacc only,\" in sacc else sacc_incl\n",
    "    print (f\"{token}{yeast}{sacc}{sacc_incl}\".rstrip(', '))\n",
    "    return (f\"{token}{yeast}{sacc}{sacc_incl}\".rstrip(', '))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "3c405944-7331-4398-b525-0dcce8f974e4",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'japonicus': ['embedding-downstream_300-static_agnostic-japonicus-512_withstop_bertstyle',\n",
      "               'embedding-species_downstream_300-static_schizosaccharomyces_japonicus-japonicus-512_withstop_bertstyle'],\n",
      " 'microarray': ['embedding-upstream_1000-static_agnostic-microarray-1024_withoutstop_bertstyle',\n",
      "                'embedding-species_upstream_1000-static_many_species-microarray-1024_withoutstop_bertstyle'],\n",
      " 'microarray_downstream': ['embedding-downstream_300-static_agnostic-microarray_downstream-512_withstop_bertstyle',\n",
      "                           'embedding-species_downstream_300-static_many_species-microarray_downstream-512_withstop_bertstyle'],\n",
      " 'pombe': ['embedding-species_downstream_300-static_schizosaccharomyces_pombe-pombe-512_withstop_bertstyle',\n",
      "           'embedding-downstream_300-static_agnostic-pombe-512_withstop_bertstyle'],\n",
      " 'scer_downstream_fixedlen': ['embedding-downstream_300-static_agnostic-scer_downstream_fixedlen-512_withstop_bertstyle',\n",
      "                              'embedding-species_downstream_300-static_candida_glabrata-scer_downstream_fixedlen-512_withstop_bertstyle'],\n",
      " 'scer_upstream': ['embedding-upstream_1000-static_agnostic-scer_upstream-1024_withoutstop_bertstyle',\n",
      "                   'embedding-species_upstream_1000-static_kazachstania_africana_cbs_2517_gca_000304475-scer_upstream-1024_withoutstop_bertstyle'],\n",
      " 'segal': ['embedding-downstream_300-static_agnostic-segal-512_withstop_bertstyle',\n",
      "           'embedding-species_downstream_300-static_candida_glabrata-segal-512_withstop_bertstyle'],\n",
      " 'segal_5prime': ['embedding-species_upstream_1000-static_kazachstania_africana_cbs_2517_gca_000304475-segal_5prime-1024_withoutstop_bertstyle',\n",
      "                  'embedding-upstream_1000-static_agnostic-segal_5prime-1024_withoutstop_bertstyle'],\n",
      " 'yamanishi': ['embedding-downstream_300-static_agnostic-yamanishi-512_withstop_bertstyle',\n",
      "               'embedding-species_downstream_300-static_candida_glabrata-yamanishi-512_withstop_bertstyle']}\n"
     ]
    }
   ],
   "source": [
    "import pprint\n",
    "pp = pprint.PrettyPrinter(depth=4)\n",
    "pp.pprint(dataset_and_model_map)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "31d167e6-297f-46d8-9124-d4a1aadad3c2",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def get_model_name(path):\n",
    "    token = \" \".join(path.split(\"-\")[2].split(\"_\")[1:])\n",
    "    model_name  = path.split(\"-\")[1]\n",
    "    yeast = \"yeast only, \" if \"yeast\" in path else \"\" \n",
    "    sacc = \"sacc only, \" if \"_sacc_incl\" in path else \"\"\n",
    "    token = f\"Aware:{token}, \" if not 'agnostic' in token else f\"Agnostic, \"\n",
    "    print (f\"{token}\".rstrip(', '))\n",
    "    return (f\"{token}\".rstrip(', '))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "79af4b26-bd54-4b81-95d4-8f9a8bbe355d",
   "metadata": {},
   "source": [
    "# S. cerevisiae Half Life"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8d7e7884-5027-4126-ac7d-05d120739b54",
   "metadata": {},
   "source": [
    "## Prepare data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "33d31d49-669e-42f2-b3c8-eb4bf8d4062b",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Agnostic\n",
      "Aware:candida glabrata\n",
      "2-mer counts\n",
      "3-mer counts\n",
      "4-mer counts\n",
      "5-mer counts\n",
      "Cheng Features\n",
      "Agnostic\n",
      "Aware:candida glabrata\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>r2</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>model</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2-mer counts</th>\n",
       "      <td>0.577900</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3-mer counts</th>\n",
       "      <td>0.580361</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4-mer counts</th>\n",
       "      <td>0.569335</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5-mer counts</th>\n",
       "      <td>0.532131</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Agnostic</th>\n",
       "      <td>0.578197</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Aware:candida glabrata</th>\n",
       "      <td>0.592134</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Cheng Features</th>\n",
       "      <td>0.586421</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                              r2\n",
       "model                           \n",
       "2-mer counts            0.577900\n",
       "3-mer counts            0.580361\n",
       "4-mer counts            0.569335\n",
       "5-mer counts            0.532131\n",
       "Agnostic                0.578197\n",
       "Aware:candida glabrata  0.592134\n",
       "Cheng Features          0.586421"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WilcoxonResult(statistic=4674612.0, pvalue=0.09499300181673248)\n",
      "WilcoxonResult(statistic=4788107.0, pvalue=0.7510424021344188)\n"
     ]
    }
   ],
   "source": [
    "def scer_half_life(path_list):\n",
    "    #Data Prep\n",
    "    halflife_df = pd.read_csv(\"./data/Downstream_Targets/Scer_half_life.csv\")\n",
    "    codon_vars = ['TTC','TTA', 'TTG', 'TCT', 'TCC', 'TCA', 'TCG', 'TAT', 'TAC', 'TGT', 'TGC',\n",
    "       'TGG', 'CTT', 'CTC', 'CTA', 'CTG', 'CCT', 'CCC', 'CCA', 'CCG', 'CAT',\n",
    "       'CAC', 'CAA', 'CAG', 'CGT', 'CGC', 'CGA', 'CGG', 'ATT', 'ATC', 'ATA',\n",
    "       'ATG', 'ACT', 'ACC', 'ACA', 'ACG', 'AAT', 'AAC', 'AAA', 'AAG', 'AGT',\n",
    "       'AGC', 'AGA', 'AGG', 'GTT', 'GTC', 'GTA', 'GTG', 'GCT', 'GCC', 'GCA',\n",
    "       'GCG', 'GAT', 'GAC', 'GAA', 'GAG', 'GGT', 'GGC', 'GGA', 'GGG', 'TTT']\n",
    "    signif_vars = [\"UTR5_length_log\",\"GC_content_UTR5\",\"CDS_deltaG\",\"num_uAUG\",\"UTR3_length_log\",\"TGTAAATA\",\"TGCAT\",\"ATATTC\",\"TTTTTTA\"]\n",
    "    # recompute GC content\n",
    "    halflife_df['GC_content_UTR5'] = (halflife_df['UTR5_seq'].str.count('G') + halflife_df['UTR5_seq'].str.count('C'))/halflife_df['UTR5_seq'].str.len()\n",
    "    halflife_df['GC_content_UTR3'] = (halflife_df['UTR3_seq'].str.count('G') + halflife_df['UTR3_seq'].str.count('C'))/halflife_df['UTR3_seq'].str.len()\n",
    "    halflife_df['GC_content_CDS'] = (halflife_df['CDS_seq'].str.count('G') + halflife_df['CDS_seq'].str.count('C'))/halflife_df['CDS_seq'].str.len()\n",
    "    # make uAUG an indicator\n",
    "    halflife_df['num_uAUG'] = halflife_df['num_uAUG'] > 0\n",
    "    # make codons into ratios\n",
    "    for codon in codon_vars[:-1]:\n",
    "        halflife_df[codon] = halflife_df[codon]/halflife_df['CDS_length']\n",
    "    # finally log the length\n",
    "    halflife_df['UTR5_length_log'] = np.log2(halflife_df['UTR5_length'])\n",
    "    halflife_df['UTR3_length_log'] = np.log2(halflife_df['UTR3_length'])\n",
    "    halflife_df['CDS_length_log'] = np.log2(halflife_df['CDS_length'])\n",
    "    halflife_df[\"log_hl\"] = np.log2(halflife_df[\"hlt.wt\"])\n",
    "    three_sequences = pd.read_parquet(\"./data/Sequences/Annotation/Sequences/saccharomyces_cerevisiae/saccharomyces_cerevisiae_three_prime.parquet\")\n",
    "    \n",
    "    scer_emb_dict = {} \n",
    "    for path in path_list:\n",
    "        tmp = np.load(f\"{project_path}/{path}/embeddings.npy\")            \n",
    "        scer_emb_dict[get_model_name(path)] = [tmp[i] for i in range(tmp.shape[0])]\n",
    "    \n",
    "    for model_name, emb in scer_emb_dict.items():\n",
    "        three_sequences[model_name] = emb\n",
    "    three_sequences.drop_duplicates(subset=['gene_id'], keep='first', inplace=True)\n",
    "\n",
    "    halflife_df = halflife_df.merge(three_sequences.rename(columns={\"gene_id\":\"genename\"}),on=\"genename\")\n",
    "   \n",
    "    y_obs = np.array(halflife_df[\"log_hl\"])\n",
    "\n",
    "    X_base = np.array(halflife_df[codon_vars[:-1] + signif_vars[:-4]])\n",
    "\n",
    "    X_jun = np.array(halflife_df[codon_vars[:-1] + signif_vars])\n",
    "    \n",
    "    X_models = {}\n",
    "    for model in scer_emb_dict.keys():\n",
    "        X_models[model] = np.concatenate([X_base, np.stack(halflife_df[model])],axis=1)\n",
    "        \n",
    "    X_gc = np.array(halflife_df[codon_vars[:-1] + signif_vars[:-4] + ['GC_content_UTR3']])\n",
    "    X_2mer = np.concatenate([X_base, np.stack(halflife_df['UTR3_seq'].apply(lambda x: kmerizer2.kmerize(x)))],axis=1)\n",
    "    X_3mer = np.concatenate([X_base, np.stack(halflife_df['UTR3_seq'].apply(lambda x: kmerizer3.kmerize(x)))],axis=1)\n",
    "    X_4mer = np.concatenate([X_base, np.stack(halflife_df['UTR3_seq'].apply(lambda x: kmerizer4.kmerize(x)))],axis=1)\n",
    "    X_5mer = np.concatenate([X_base, np.stack(halflife_df['UTR3_seq'].apply(lambda x: kmerizer5.kmerize(x)))],axis=1)\n",
    "    X_6mer = np.concatenate([X_base, np.stack(halflife_df['UTR3_seq'].apply(lambda x: kmerizer6.kmerize(x)))],axis=1)\n",
    "    \n",
    "    #Predict Half Life\n",
    "    \n",
    "    data_matrices = {\n",
    "                 \"2-mer counts\":X_2mer, \n",
    "                 \"3-mer counts\":X_3mer, \n",
    "                 \"4-mer counts\":X_4mer,\n",
    "                 \"5-mer counts\":X_5mer, \n",
    "                 \"Cheng Features\":X_jun}\n",
    "    data_matrices.update(X_models)\n",
    "\n",
    "    rows = []\n",
    "    test_predictions = []\n",
    "    total_len = X_base.shape[0]\n",
    "    n_splits = 10\n",
    "\n",
    "    kfold = sklearn.model_selection.KFold(n_splits=n_splits, shuffle=True, random_state=42)\n",
    "\n",
    "    for key in data_matrices:\n",
    "        print(key)\n",
    "        y = y_obs\n",
    "        X = data_matrices[key]\n",
    "\n",
    "        y_pred_list = []\n",
    "        y_true_list = []\n",
    "        for i, (train_index, test_index) in enumerate(kfold.split(X, y)):\n",
    "            pipe = pipeline.make_pipeline(sklearn.preprocessing.StandardScaler(), sklearn.linear_model.RidgeCV(cv=5, alphas=[0.1, 1, 10, 100, 1000,10000]))\n",
    "            X_train = X[train_index]\n",
    "            y_train = y[train_index]\n",
    "            X_test = X[test_index]\n",
    "            y_test = y[test_index]\n",
    "            pipe.fit(X_train,y_train)\n",
    "            y_pred = pipe.predict(X_test)\n",
    "            r2_test = pearson_r2_metric(y_test, y_pred)\n",
    "            # save results\n",
    "            y_pred_list.append(y_pred)\n",
    "            y_true_list.append(y_test)\n",
    "            rows.append({\"model\":key, \"r2\":r2_test})\n",
    "        test_predictions.append(pd.DataFrame({\"model\":key, \n",
    "                                 \"pred\": np.concatenate(y_pred_list), \n",
    "                                 \"true\": np.concatenate(y_true_list)}))\n",
    "\n",
    "    metrics_scer = pd.DataFrame(rows)\n",
    "    pred_scer = pd.concat(test_predictions)\n",
    "    pred_scer[\"residual\"] = (pred_scer[\"true\"] - pred_scer[\"pred\"])**2\n",
    "    display(metrics_scer.groupby('model').median())\n",
    "    print(scipy.stats.wilcoxon(pred_scer.query('model == \"Aware:candida glabrata\"')[\"residual\"], pred_scer.query('model == \"Agnostic\"')[\"residual\"]))\n",
    "    print(scipy.stats.wilcoxon(pred_scer.query('model == \"Aware:candida glabrata\"')[\"residual\"], pred_scer.query('model == \"Cheng Features\"')[\"residual\"]))\n",
    "    return metrics_scer, pred_scer\n",
    "    \n",
    "metrics_scer, pred_scer = scer_half_life(dataset_and_model_map[\"scer_downstream_fixedlen\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5d8bb887-5d60-47ce-8118-932bc89a143c",
   "metadata": {},
   "source": [
    "# S. pombe Half Life"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "0536546d-9ecc-4cd8-b554-4391f33f54ad",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def pombe_half_life(path_list):\n",
    "    #Prepare data\n",
    "    halflife_pombe_df = pd.read_csv(\"./data/Downstream_Targets/Pombe_half_life.csv\")\n",
    "    codon_vars_pombe = ['TTC','TTA', 'TTG', 'TCT', 'TCC', 'TCA', 'TCG', 'TAT', 'TAC', 'TGT', 'TGC',\n",
    "       'TGG', 'CTT', 'CTC', 'CTA', 'CTG', 'CCT', 'CCC', 'CCA', 'CCG', 'CAT',\n",
    "       'CAC', 'CAA', 'CAG', 'CGT', 'CGC', 'CGA', 'CGG', 'ATT', 'ATC', 'ATA',\n",
    "       'ATG', 'ACT', 'ACC', 'ACA', 'ACG', 'AAT', 'AAC', 'AAA', 'AAG', 'AGT',\n",
    "       'AGC', 'AGA', 'AGG', 'GTT', 'GTC', 'GTA', 'GTG', 'GCT', 'GCC', 'GCA',\n",
    "       'GCG', 'GAT', 'GAC', 'GAA', 'GAG', 'GGT', 'GGC', 'GGA', 'GGG', 'TTT']\n",
    "\n",
    "    signif_vars_pombe = ['UTR3_length_log','UTR5_length_log','CDS_length_log',\n",
    "               \"stopPlus1\",\n",
    "               \"num_uAUG\",\"CAACCA\",\"ACCAAC\",\n",
    "               \"GC_content_UTR5\",\"UTR5_deltaG\",\"CDS_deltaG\",\"GC_content_CDS\",\n",
    "               \"UTR3_deltaG\",\"GC_content_UTR3\",\"TATTTAT\",\"TTAATGA\",\"ACTAAT\"]\n",
    "    # recompute GC content\n",
    "    halflife_pombe_df['GC_content_UTR5'] = (halflife_pombe_df['UTR5_seq'].str.count('G') + halflife_pombe_df['UTR5_seq'].str.count('C'))/halflife_pombe_df['UTR5_seq'].str.len()\n",
    "    halflife_pombe_df['GC_content_UTR3'] = (halflife_pombe_df['UTR3_seq'].str.count('G') + halflife_pombe_df['UTR3_seq'].str.count('C'))/halflife_pombe_df['UTR3_seq'].str.len()\n",
    "    #halflife_pombe_df['GC_content_CDS'] = (halflife_pombe_df['CDS_seq'].str.count('G') + halflife_pombe_df['CDS_seq'].str.count('C'))/halflife_pombe_df['CDS_seq'].str.len()\n",
    "    # make uAUG an indicator\n",
    "    halflife_pombe_df['num_uAUG'] = halflife_pombe_df['num_uAUG'] > 0\n",
    "    # make codons into ratios\n",
    "    for codon in codon_vars_pombe[:-1]:\n",
    "        halflife_pombe_df[codon] = halflife_pombe_df[codon]/halflife_pombe_df['CDS_length']\n",
    "    # finally log the length\n",
    "    halflife_pombe_df['UTR5_length_log'] = np.log2(halflife_pombe_df['UTR5.length'])\n",
    "    halflife_pombe_df['UTR3_length_log'] = np.log2(halflife_pombe_df['UTR3.length'])\n",
    "    halflife_pombe_df['CDS_length_log'] = np.log2(halflife_pombe_df['CDS_length'])\n",
    "    halflife_pombe_df[\"log_hl\"] = np.log2(halflife_pombe_df[\"half.life\"])\n",
    "    ols_hl_pombe_full = (statsmodels.regression.linear_model.OLS\n",
    "                  .from_formula('log_hl ~ {}'.format(\"+\".join(codon_vars_pombe + signif_vars_pombe)),\n",
    "                                data=halflife_pombe_df))#.query('UTR5_length >= 20 and UTR3_length >= 20')))\n",
    "\n",
    "    res_hl_pombe_full = ols_hl_pombe_full.fit()\n",
    "    \n",
    "    three_sequences = pd.read_csv(\"./data/Sequences/Annotation/Sequences/schizosaccharomyces_pombe/schizosaccharomyces_pombe_three_prime_remapped.tsv\",sep=\"\\t\")\n",
    "    \n",
    "    pombe_emb_dict = {} \n",
    "    for path in path_list:\n",
    "        tmp = np.load(f\"{project_path}/{path}/embeddings.npy\")            \n",
    "        pombe_emb_dict[get_model_name(path)] = [tmp[i] for i in range(tmp.shape[0])]\n",
    "    \n",
    "    for model_name, emb in pombe_emb_dict.items():\n",
    "        three_sequences[model_name] = emb\n",
    "    three_sequences.drop_duplicates(subset=['gene_id'], keep='first', inplace=True)\n",
    "\n",
    "\n",
    "    halflife_pombe_df = halflife_pombe_df.merge(three_sequences[[\"map.ids\"] +list(pombe_emb_dict.keys())],on=\"map.ids\")\n",
    "    y_obs = np.array(halflife_pombe_df[\"log_hl\"])\n",
    "\n",
    "    X_base = np.asarray(patsy.dmatrix('~ -1 + {}'.format(\"+\".join(codon_vars_pombe[:-1] + signif_vars_pombe[:-5])), \n",
    "                  data=halflife_pombe_df))\n",
    "\n",
    "    X_jun = np.concatenate([X_base, np.array(halflife_pombe_df[signif_vars_pombe[-5:]])],axis=1)\n",
    "    #X_emb = np.concatenate([X_base, emb_pombe],axis=1)\n",
    "    X_gc = np.concatenate([X_base, np.array(halflife_pombe_df[['GC_content_UTR3']])],axis=1)\n",
    "    \n",
    "    X_models = {}\n",
    "    for model in pombe_emb_dict.keys():\n",
    "        X_models[model] = np.concatenate([X_base, np.stack(halflife_pombe_df[model])],axis=1)\n",
    "\n",
    "    X_2mer = np.concatenate([X_base, np.stack(halflife_pombe_df['UTR3_seq'].apply(lambda x: kmerizer2.kmerize(x)))],axis=1)\n",
    "    X_3mer = np.concatenate([X_base, np.stack(halflife_pombe_df['UTR3_seq'].apply(lambda x: kmerizer3.kmerize(x)))],axis=1)\n",
    "    X_4mer = np.concatenate([X_base, np.stack(halflife_pombe_df['UTR3_seq'].apply(lambda x: kmerizer4.kmerize(x)))],axis=1)\n",
    "    X_5mer = np.concatenate([X_base, np.stack(halflife_pombe_df['UTR3_seq'].apply(lambda x: kmerizer5.kmerize(x)))],axis=1)\n",
    "    X_6mer = np.concatenate([X_base, np.stack(halflife_pombe_df['UTR3_seq'].apply(lambda x: kmerizer6.kmerize(x)))],axis=1)\n",
    "    \n",
    "    #Pred Half Life\n",
    "    \n",
    "    data_matrices = {\"2-mer counts\":X_2mer, \n",
    "                 \"3-mer counts\":X_3mer, \n",
    "                 \"4-mer counts\":X_4mer,\n",
    "                 \"5-mer counts\":X_5mer, \n",
    "                 \"Eser Features\":X_jun}\n",
    "    data_matrices.update(X_models)\n",
    "    rows = []\n",
    "    test_predictions = []\n",
    "    total_len = X_base.shape[0]\n",
    "    n_splits = 10\n",
    "\n",
    "    kfold = sklearn.model_selection.KFold(n_splits=n_splits, shuffle=True, random_state=42)\n",
    "\n",
    "    for key in data_matrices:\n",
    "        print(key)\n",
    "        y = y_obs\n",
    "        X = data_matrices[key]\n",
    "\n",
    "        y_pred_list = []\n",
    "        y_true_list = []\n",
    "        for i, (train_index, test_index) in enumerate(kfold.split(X, y)):\n",
    "            pipe = pipeline.make_pipeline(sklearn.preprocessing.StandardScaler(), sklearn.linear_model.RidgeCV(cv=5, alphas=[0.1, 1, 10, 100, 1000,10000]))\n",
    "            X_train = X[train_index]\n",
    "            y_train = y[train_index]\n",
    "            X_test = X[test_index]\n",
    "            y_test = y[test_index]\n",
    "            pipe.fit(X_train,y_train)\n",
    "            y_pred = pipe.predict(X_test)\n",
    "            r2_test = pearson_r2_metric(y_test, y_pred)\n",
    "            # save results\n",
    "            y_pred_list.append(y_pred)\n",
    "            y_true_list.append(y_test)\n",
    "            rows.append({\"model\":key, \"r2\":r2_test})\n",
    "        test_predictions.append(pd.DataFrame({\"model\":key, \n",
    "                                 \"pred\": np.concatenate(y_pred_list), \n",
    "                                 \"true\": np.concatenate(y_true_list)}))\n",
    "\n",
    "    metrics_pombe = pd.DataFrame(rows)\n",
    "    pred_pombe = pd.concat(test_predictions)\n",
    "    pred_pombe[\"residual\"] = (pred_pombe[\"true\"] - pred_pombe[\"pred\"])**2\n",
    "    display(metrics_pombe.groupby('model').agg(\"median\"))\n",
    "    return metrics_pombe, pred_pombe"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "c9834225-b156-4a05-9a6b-559b06064c02",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Aware:schizosaccharomyces pombe\n",
      "Agnostic\n",
      "2-mer counts\n",
      "3-mer counts\n",
      "4-mer counts\n",
      "5-mer counts\n",
      "Eser Features\n",
      "Aware:schizosaccharomyces pombe\n",
      "Agnostic\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>r2</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>model</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2-mer counts</th>\n",
       "      <td>0.421239</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3-mer counts</th>\n",
       "      <td>0.432528</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4-mer counts</th>\n",
       "      <td>0.450010</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5-mer counts</th>\n",
       "      <td>0.400181</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Agnostic</th>\n",
       "      <td>0.479087</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Aware:schizosaccharomyces pombe</th>\n",
       "      <td>0.494266</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Eser Features</th>\n",
       "      <td>0.434660</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                       r2\n",
       "model                                    \n",
       "2-mer counts                     0.421239\n",
       "3-mer counts                     0.432528\n",
       "4-mer counts                     0.450010\n",
       "5-mer counts                     0.400181\n",
       "Agnostic                         0.479087\n",
       "Aware:schizosaccharomyces pombe  0.494266\n",
       "Eser Features                    0.434660"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "metrics_pombe, pred_pombe = pombe_half_life(dataset_and_model_map[\"pombe\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "f2604591-d07a-449b-9213-b287afdae5e2",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "WilcoxonResult(statistic=2949429.0, pvalue=2.055200150281855e-13)"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scipy.stats.wilcoxon(pred_pombe.query('model == \"Aware:schizosaccharomyces pombe\"')[\"residual\"], pred_pombe.query('model == \"Eser Features\"')[\"residual\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9b681b38-e755-47f8-851c-677683f22ace",
   "metadata": {},
   "source": [
    "# 3' UTR MPRA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "1ea17b57-0f68-490d-b8d0-8c0fd134d6f4",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def segal_mpra(path_list):\n",
    "    mpra_df = pd.read_csv(\"./data/Downstream_Targets/segal_2015.tsv\",sep=\"\\t\").dropna().reset_index(drop=True).reset_index()\n",
    "\n",
    "    mpra_df['TGTAAATA'] = mpra_df['Oligo Sequence'].str.count('TGTAAATA')\n",
    "    mpra_df['TGCAT'] = mpra_df['Oligo Sequence'].str.count('TGCAT')\n",
    "    mpra_df['ATATTC'] = mpra_df['Oligo Sequence'].str.count('ATATTC')\n",
    "    mpra_df['TTTTTTA'] = mpra_df['Oligo Sequence'].str.count('TTTTTTA')\n",
    "    mpra_df['GC_content_UTR3'] = (mpra_df['Oligo Sequence'].str.count('G') + mpra_df['Oligo Sequence'].str.count('C'))/mpra_df['Oligo Sequence'].str.len()\n",
    "\n",
    "    def replacer(x):\n",
    "        x = (x.replace(\"native_\",\"\")\n",
    "                .replace(\"pad_\",\"\")\n",
    "                .replace(\"Lib_pre_\",\"\")\n",
    "                .replace(\"term_full_\",\"\")\n",
    "                .replace(\"term_null_\",\"\")\n",
    "                .replace(\"term_ins_\",\"\")\n",
    "                .replace(\"lib_suff_nag_\",\"\")\n",
    "                .replace(\"ib_suff_osz_\",\"\")\n",
    "                .replace(\"stop_codon_\",\"\")\n",
    "                .replace(\"canon_term_neg_\",\"\")\n",
    "                .replace(\"canon_term_pos_\",\"\")\n",
    "                .replace(\"_no_ins\",\"\")\n",
    "                .replace(\"_three_ins\",\"\")\n",
    "                .replace(\"_four_ins\",\"\")\n",
    "                .replace(\"_two_ins\",\"\")\n",
    "                .replace(\"_one_ins\",\"\")\n",
    "                .replace(\"_one\",\"\")\n",
    "                .replace(\"_synth_insert_pad\",\"\")\n",
    "                .replace(\"_synth_t\",\"\")\n",
    "                .replace(\"_512_mut\",\"\")\n",
    "                .replace(\"_wt\",\"\")\n",
    "               )\n",
    "        if x.startswith('l'):\n",
    "            return x[1:]\n",
    "        else:\n",
    "            return x\n",
    "\n",
    "    mpra_df['block'] = [replacer(x.split(\"#\")[1].split(\"context_\")[-1]) for x in mpra_df[\"Description\"]]\n",
    "\n",
    "    y_obs = np.array(np.log2(mpra_df[\"Expression\"]))\n",
    "    groups = np.array(mpra_df['block'])\n",
    "\n",
    "    X_2mer = np.stack(mpra_df['Oligo Sequence'].apply(lambda x: kmerizer2.kmerize(x)))\n",
    "    X_3mer = np.stack(mpra_df['Oligo Sequence'].apply(lambda x: kmerizer3.kmerize(x)))\n",
    "    X_4mer = np.stack(mpra_df['Oligo Sequence'].apply(lambda x: kmerizer4.kmerize(x)))\n",
    "    X_5mer = np.stack(mpra_df['Oligo Sequence'].apply(lambda x: kmerizer5.kmerize(x)))\n",
    "    X_6mer = np.stack(mpra_df['Oligo Sequence'].apply(lambda x: kmerizer6.kmerize(x)))\n",
    "    \n",
    "    segal_emb_dict = {} \n",
    "    for path in path_list:\n",
    "        tmp = np.load(f\"{project_path}/{path}/embeddings.npy\")            \n",
    "        segal_emb_dict[get_model_name(path)] = np.array([tmp[i] for i in range(tmp.shape[0])])  \n",
    "    \n",
    "    data_matrices = {\"2-mer counts\":X_2mer, \n",
    "                     \"3-mer counts\":X_3mer, \n",
    "                     \"4-mer counts\":X_4mer,\n",
    "                     \"5-mer counts\":X_5mer,\n",
    "                    }\n",
    "    data_matrices.update(segal_emb_dict)\n",
    "    rows = []\n",
    "    df_list = []\n",
    "    np.random.seed(42)\n",
    "    test_predictions = []\n",
    "\n",
    "    group_kfold = sklearn.model_selection.GroupKFold(n_splits=10)\n",
    "\n",
    "    for key in data_matrices:\n",
    "        print(key)\n",
    "        y = y_obs\n",
    "        X = data_matrices[key]\n",
    "\n",
    "        y_pred_list = []\n",
    "        y_true_list = []\n",
    "        for i, (train_index, test_index) in enumerate(group_kfold.split(X, y, groups)):\n",
    "            pipe = pipeline.make_pipeline(sklearn.preprocessing.StandardScaler(), sklearn.linear_model.RidgeCV(cv=3, alphas=[10, 100, 1000])) #limit runtime\n",
    "            X_train = X[train_index]\n",
    "            y_train = y[train_index]\n",
    "            X_test = X[test_index]\n",
    "            y_test = y[test_index]\n",
    "            pipe.fit(X_train,y_train)\n",
    "            y_pred = pipe.predict(X_test)\n",
    "            r2_test = pearson_r2_metric(y_test, y_pred)\n",
    "            # save results\n",
    "            y_pred_list.append(y_pred)\n",
    "            y_true_list.append(y_test)\n",
    "            rows.append({\"model\":key, \"r2\":r2_test})\n",
    "        test_predictions.append(pd.DataFrame({\"model\":key, \n",
    "                                 \"pred\": np.concatenate(y_pred_list), \n",
    "                                 \"true\": np.concatenate(y_true_list)}))\n",
    "\n",
    "    metrics_mpra = pd.DataFrame(rows)\n",
    "    pred_mpra = pd.concat(test_predictions)\n",
    "    pred_mpra[\"residual\"] = (pred_mpra[\"true\"] - pred_mpra[\"pred\"])**2\n",
    "    bert_adn_preds = pred_mpra.query(\"model == 'Aware:candida glabrata'\")\n",
    "    bert_adn_preds_agnostic = pred_mpra.query(\"model == 'Agnostic'\")\n",
    "    bert_adn_preds[\"pred_agnostic\"] = bert_adn_preds_agnostic[\"pred\"]\n",
    "    test_indices = np.concatenate([test_index for i, (train_index, test_index) in enumerate(group_kfold.split(X, y, groups))])\n",
    "    bert_adn_preds[\"index\"] = test_indices\n",
    "    bert_adn_preds = bert_adn_preds.merge(mpra_df, on=\"index\")\n",
    "    return metrics_mpra, bert_adn_preds, pred_mpra"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "a2229014-fc14-4ef1-bde9-5ae25f2e1d9b",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Agnostic\n",
      "Aware:candida glabrata\n",
      "2-mer counts\n",
      "3-mer counts\n",
      "4-mer counts\n",
      "5-mer counts\n",
      "Agnostic\n",
      "Aware:candida glabrata\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/scratch/tmp/hingerl/ipykernel_891843/2897999629.py:96: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "/scratch/tmp/hingerl/ipykernel_891843/2897999629.py:98: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>r2</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>model</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3-mer counts</th>\n",
       "      <td>0.276111</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4-mer counts</th>\n",
       "      <td>0.278299</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2-mer counts</th>\n",
       "      <td>0.303907</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5-mer counts</th>\n",
       "      <td>0.336683</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Agnostic</th>\n",
       "      <td>0.531816</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Aware:candida glabrata</th>\n",
       "      <td>0.704453</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                              r2\n",
       "model                           \n",
       "3-mer counts            0.276111\n",
       "4-mer counts            0.278299\n",
       "2-mer counts            0.303907\n",
       "5-mer counts            0.336683\n",
       "Agnostic                0.531816\n",
       "Aware:candida glabrata  0.704453"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics_mpra, bert_adn_preds, pred_mpra = segal_mpra(dataset_and_model_map[\"segal\"])\n",
    "metrics_mpra.groupby('model').agg(\"median\").sort_values('r2')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "bf7c137d-f0e3-47cd-9508-35f4954a0173",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "WilcoxonResult(statistic=27866599.0, pvalue=0.0)"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scipy.stats.wilcoxon(pred_mpra.query('model == \"Aware:candida glabrata\"')[\"residual\"], pred_mpra.query('model == \"Agnostic\"')[\"residual\"])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "b739cfe0-ce23-45dc-9141-db3d3dbe0d77",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "def analyze_fc(bert_adn_preds,gene):\n",
    "    plot_df = bert_adn_preds.loc[bert_adn_preds[\"block\"].str.contains(gene)]\n",
    "    wt_expr_true = plot_df.loc[plot_df[\"Description\"].str.contains(\"Native 3 UTRs\")].iloc[0][\"true\"]\n",
    "    wt_expr_pred = plot_df.loc[plot_df[\"Description\"].str.contains(\"Native 3 UTRs\")].iloc[0][\"pred\"]\n",
    "    wt_expr_pred_agnostic = plot_df.loc[plot_df[\"Description\"].str.contains(\"Native 3 UTRs\")].iloc[0][\"pred_agnostic\"]\n",
    "\n",
    "    plot_df = plot_df.loc[plot_df[\"Description\"].str.contains(\"Scanning mutagenesis\")]\n",
    "    assert len(set(plot_df[\"block\"])) == 1\n",
    "    plot_df[\"dist\"] = plot_df[\"Description\"].apply(lambda x: int(x.split(\"#\")[2].split(\"=\")[-1]))\n",
    "    plot_df[\"dist\"] = 140 - plot_df[\"dist\"]\n",
    "    plot_df[\"logfc_true\"] = plot_df[\"true\"] - wt_expr_true#np.log2(plot_df[\"true\"]/wt_expr_true)\n",
    "    plot_df[\"logfc_pred\"] = plot_df[\"pred\"] - wt_expr_pred#np.log2(plot_df[\"pred\"]/wt_expr_pred)\n",
    "    plot_df[\"logfc_pred_agnostic\"] = plot_df[\"pred_agnostic\"] - wt_expr_pred_agnostic#np.log2(plot_df[\"pred_agnostic\"]/wt_expr_pred_agnostic)\n",
    "    # set impossible fold changes to zero\n",
    "    #plot_df[\"logfc_pred\"] = plot_df[\"logfc_pred\"].fillna(0)\n",
    "    #plot_df[\"logfc_pred_agnostic\"] = plot_df[\"logfc_pred_agnostic\"].fillna(0)\n",
    "\n",
    "    plot_df = (plot_df[[\"dist\",\"logfc_pred\",\"logfc_true\",\"logfc_pred_agnostic\"]]\n",
    "               .rename(columns={\"logfc_pred\":\"Species LM\",\"logfc_pred_agnostic\":\"Agnostic LM\",\"logfc_true\":\"Observed\"})\n",
    "               .groupby('dist').mean().reset_index().melt(\"dist\"))\n",
    "    plot_df[\"variable\"] = pd.Categorical(plot_df[\"variable\"], ordered=True, categories=['Observed',\"Agnostic LM\", 'Species LM'])\n",
    "\n",
    "    p = (p9.ggplot(data=plot_df, mapping=p9.aes(x=\"dist\",y=\"value\",color=\"variable\",group=\"variable\")) \n",
    "     + p9.geom_point(size = 1) \n",
    "     + p9.geom_line(size = 1)\n",
    "     + p9.scale_color_manual(values=[\"mediumblue\",\"salmon\",\"crimson\"])\n",
    "    + p9.theme_minimal()\n",
    "         +p9.theme(axis_title=p9.element_text(size=10),legend_position = (0.70,0.248), legend_direction=\"vertical\", legend_background=p9.element_blank(), legend_key_size=9,\n",
    "            legend_title=p9.element_text(size=9),legend_text = p9.element_text(size=9),\n",
    "             )\n",
    "     #+ p9.geom_smooth(method=\"lm\")\n",
    "     + p9.labs(x=\"Distance to stop codon\",y=\"log$_2$ Fold Change w.r.t. WT\",color=\"\")\n",
    "    )\n",
    "    return plot_df, p\n",
    "plot_df, p = analyze_fc(bert_adn_preds, \"YDR131C\")\n",
    "print (p)\n",
    "#p.save(\"Downstream/Graphics/Paper/Downstream_Segal_FC_YDR131C.svg\", width=3, height=2.5)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bf6baa86-cd16-4c0d-84f3-66a02c934a8e",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "c8d1c805-d43e-4805-bb8b-cf1e1ed1d637",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "#%%capture\n",
    "scanned_genes = set([x.split(\"#\")[1].split(\"context_native_\")[-1] for x in bert_adn_preds.loc[bert_adn_preds[\"Description\"].str.contains(\"Scanning mutagenesis\")][\"Description\"]])\n",
    "plot_df_list = []\n",
    "for gene in scanned_genes:\n",
    "    try:\n",
    "        plot_df = analyze_fc(bert_adn_preds, gene)[0]\n",
    "        plot_df[\"gene\"] = gene\n",
    "        plot_df_list.append(plot_df)\n",
    "    except IndexError:\n",
    "        # genes where we have no wt\n",
    "        #print(gene)\n",
    "        continue"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "b47e1cf0-1ef0-414f-b1bc-0b688705aa1c",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "plot_df = pd.concat(plot_df_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "25604b3b-7369-498a-9ecb-c433918d81ce",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<ggplot: (1409169869041)>"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "plot_df_full = pd.concat([plot_df.query('variable == \"Species LM\"')[[\"gene\",\"value\"]].rename(columns={\"value\":\"pred\"}).reset_index(drop=True), \n",
    "                          plot_df.query('variable == \"Observed\"')[[\"value\"]].rename(columns={\"value\":\"obs\"}).reset_index(drop=True)],axis=1)\n",
    "\n",
    "p = (p9.ggplot(data=plot_df_full, mapping=p9.aes(x=\"pred\",y=\"obs\")) \n",
    " #+ p9.geom_point(size=1,alpha=0.15)\n",
    " + p9.geom_point(size=2,alpha=0.1)\n",
    " + p9.geom_smooth(method=\"lm\")\n",
    " + p9.theme_minimal()\n",
    " + p9.labs(x=\"Predicted log2 FC (Species)\",y=\"Observed log2 FC\")\n",
    ")\n",
    "p"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "46903ecf-0a03-4205-9344-2f1c4c988e92",
   "metadata": {
    "tags": []
   },
   "source": [
    "# Yeast Expression (Zrimec)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "3ad97c27-15e7-4735-98e9-39b6c7c53d46",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "downstream_path_list = dataset_and_model_map[\"scer_downstream_fixedlen\"]\n",
    "upstream_path_list = dataset_and_model_map[\"scer_upstream\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "67f7321a-f43c-40f0-85ac-d440c0188e01",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "expr_df = pd.read_csv(\"./data/Downstream_Targets/scerevisiae_tpm_full_cut5.csv\").query('TPM >= 5')\n",
    "\n",
    "# apply box-cox transform with lambda = 0.22\n",
    "expr_df[\"boxcox_TPM\"] = scipy.stats.boxcox(expr_df[\"TPM\"], lmbda=0.22)\n",
    "# regress out gene length\n",
    "gene_lengths = np.array(expr_df[\"gene_len\"]).reshape(-1, 1)\n",
    "expr_df[\"residual_boxcox_TPM\"] = expr_df[\"boxcox_TPM\"] - sklearn.linear_model.LinearRegression().fit(y=expr_df[\"boxcox_TPM\"], X=gene_lengths).predict(X=gene_lengths)\n",
    "\n",
    "# take median over samples\n",
    "expr_df = expr_df.groupby('gene_id')[[\"residual_boxcox_TPM\",\"boxcox_TPM\",\"TPM\"]].median().reset_index()\n",
    "expr_df[\"log_TPM\"] = np.log10(expr_df[\"TPM\"] + 1)\n",
    "\n",
    "# get test genes\n",
    "test_genes = pd.read_csv(\"./data/Downstream_Targets/test_gene_ids.csv\")\n",
    "test_genes_set = set(test_genes[\"names_test\"])\n",
    "expr_df[\"is_test\"] = expr_df[\"gene_id\"].isin(test_genes_set)\n",
    "\n",
    "five_sequences = pd.read_parquet(\"./data/Sequences/Annotation/Sequences/saccharomyces_cerevisiae/saccharomyces_cerevisiae_five_prime.parquet\")\n",
    "cds_sequences = pd.read_parquet(\"./data/Sequences/Annotation/Sequences/saccharomyces_cerevisiae/saccharomyces_cerevisiae_cds.parquet\")\n",
    "three_sequences = pd.read_parquet(\"./data/Sequences/Annotation/Sequences/saccharomyces_cerevisiae/saccharomyces_cerevisiae_three_prime.parquet\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "3276fe7f-5d71-44d9-bbb7-af8b6100851e",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Agnostic embedding-downstream_300-static_agnostic-scer_downstream_fixedlen-512_withstop_bertstyle\n",
      "Agnostic embedding-downstream_300-static_agnostic-scer_downstream_fixedlen-512_withstop_bertstyle\n",
      "Aware:candida glabrata embedding-species_downstream_300-static_candida_glabrata-scer_downstream_fixedlen-512_withstop_bertstyle\n",
      "Aware:candida glabrata embedding-species_downstream_300-static_candida_glabrata-scer_downstream_fixedlen-512_withstop_bertstyle\n",
      "embedding-upstream_1000-static_agnostic-scer_upstream-1024_withoutstop_bertstyle\n",
      "Agnostic embedding-upstream_1000-static_agnostic-scer_upstream-1024_withoutstop_bertstyle\n",
      "Agnostic embedding-upstream_1000-static_agnostic-scer_upstream-1024_withoutstop_bertstyle\n",
      "embedding-species_upstream_1000-static_kazachstania_africana_cbs_2517_gca_000304475-scer_upstream-1024_withoutstop_bertstyle\n",
      "Aware:kazachstania africana embedding-species_upstream_1000-static_kazachstania_africana_cbs_2517_gca_000304475-scer_upstream-1024_withoutstop_bertstyle\n",
      "Aware:kazachstania africana embedding-species_upstream_1000-static_kazachstania_africana_cbs_2517_gca_000304475-scer_upstream-1024_withoutstop_bertstyle\n",
      "Upstream: Agnostic\n",
      "Upstream: Aware:kazachstania africana\n",
      "Upstream: Aware:kazachstania africana\n"
     ]
    }
   ],
   "source": [
    "three_prime_emb_dict, five_prime_emb_dict = {},{} \n",
    "def get_model_name(path):\n",
    "    token = \" \".join(path.split(\"-\")[2].split(\"_\")[1:])\n",
    "    if len(re.findall('\\d+', token)) > 0:\n",
    "        token = \" \".join(token.split(\" \")[:2])\n",
    "    model_name  = path.split(\"-\")[1]\n",
    "    token = f\"Aware:{token}, \" if not 'agnostic' in token else f\"Agnostic, \"\n",
    "    print (f\"{token}\".rstrip(', '), path)\n",
    "    return (f\"{token}\".rstrip(', '))\n",
    "\n",
    "for path in downstream_path_list:\n",
    "    tmp = np.load(f\"{project_path}/{path}/embeddings.npy\")            \n",
    "    three_prime_emb_dict[\"Downstream: \" + get_model_name(path)] = [tmp[i] for i in range(tmp.shape[0])]\n",
    "    three_sequences[\"Downstream: \" + get_model_name(path)] = [tmp[i] for i in range(tmp.shape[0])]\n",
    "three_sequences.drop_duplicates(subset=['gene_id'], keep='first', inplace=True)\n",
    "    \n",
    "for path in upstream_path_list:\n",
    "    print (path)\n",
    "    tmp = np.load(f\"{project_path}/{path}/embeddings.npy\")            \n",
    "    five_prime_emb_dict[\"Upstream: \" + get_model_name(path)] = [tmp[i] for i in range(tmp.shape[0])]\n",
    "    five_sequences[\"Upstream: \" + get_model_name(path)] = [tmp[i] for i in range(tmp.shape[0])]\n",
    "five_sequences.drop_duplicates(subset=['gene_id'], keep='first', inplace=True)\n",
    "expr_merged = expr_df.merge(five_sequences, on=\"gene_id\").merge(cds_sequences, on=\"gene_id\").merge(three_sequences, on=\"gene_id\")\n",
    "y_obs = np.array(expr_merged[\"residual_boxcox_TPM\"])\n",
    "X_models = {}\n",
    "for model in three_prime_emb_dict.keys():\n",
    "    X_models[model] = np.stack(expr_merged[model])\n",
    "for model in five_prime_emb_dict.keys():\n",
    "    print (model)\n",
    "    five_prime_model = np.stack(expr_merged[model])\n",
    "    X_models[model] = five_prime_model\n",
    "    if \"Aware\" in model:\n",
    "        print (model)\n",
    "        X_models[f\"Combined: Downstream: Aware:candida glabrata ! {model}\"] = np.concatenate([five_prime_model, np.stack(expr_merged[\"Downstream: Aware:candida glabrata\"])],axis=1)\n",
    "    else:\n",
    "        X_models[f\"Combined: Downstream: Agnostic ! {model}\"] = np.concatenate([five_prime_model, np.stack(expr_merged[\"Downstream: Agnostic\"])],axis=1)\n",
    "    #for model_three in three_prime_emb_dict.keys():\n",
    "    #    X_models[f\"Combined: {model_three} ! {model}\"] = np.concatenate([X_base, five_prime_model, np.stack(expr_merged[model_three])],axis=1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "d6285642-e39e-423f-a56d-8391b11968b1",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "#X_base = np.zeros((y_obs.shape[0],1))\n",
    "X_3mer_five = np.stack(expr_merged['five_prime_seq'].apply(lambda x: kmerizer3.kmerize(x)))\n",
    "X_4mer_five = np.stack(expr_merged['five_prime_seq'].apply(lambda x: kmerizer4.kmerize(x)))\n",
    "X_5mer_five = np.stack(expr_merged['five_prime_seq'].apply(lambda x: kmerizer5.kmerize(x)))\n",
    "\n",
    "X_3mer_three = np.stack(expr_merged['three_prime_seq'].apply(lambda x: kmerizer3.kmerize(x)))\n",
    "X_4mer_three = np.stack(expr_merged['three_prime_seq'].apply(lambda x: kmerizer4.kmerize(x)))\n",
    "X_5mer_three = np.stack(expr_merged['three_prime_seq'].apply(lambda x: kmerizer5.kmerize(x)))\n",
    "\n",
    "X_3mer_combined = np.concatenate([X_3mer_five, X_3mer_three],axis=1)\n",
    "X_4mer_combined = np.concatenate([X_4mer_five, X_4mer_three],axis=1)\n",
    "X_5mer_combined = np.concatenate([X_5mer_five, X_5mer_three],axis=1)\n",
    "\n",
    "data_matrices = {\n",
    "                 \"3-mer-fiveprime\":X_3mer_five, \n",
    "                 \"4-mer-fiveprime\":X_4mer_five,\n",
    "                 \"5-mer-fiveprime\":X_5mer_five, \n",
    "                 \"3-mer-threeprime\":X_3mer_three,\n",
    "                 \"4-mer-threeprime\":X_4mer_three,\n",
    "                 \"5-mer-threeprime\":X_5mer_three,\n",
    "                 \"3-mer-combined\":X_3mer_combined,\n",
    "                 \"4-mer-combined\":X_4mer_combined,\n",
    "                 \"5-mer-combined\":X_5mer_combined,\n",
    "}\n",
    "data_matrices.update(X_models)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "df22556e-5beb-404d-a8d2-bc63fd5a2394",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "fb8646f9-f227-4094-9bd4-91f5a8e4ae0f",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3-mer-fiveprime\n",
      "4-mer-fiveprime\n",
      "5-mer-fiveprime\n",
      "3-mer-threeprime\n",
      "4-mer-threeprime\n",
      "5-mer-threeprime\n",
      "3-mer-combined\n",
      "4-mer-combined\n",
      "5-mer-combined\n",
      "Downstream: Agnostic\n",
      "Downstream: Aware:candida glabrata\n",
      "Upstream: Agnostic\n",
      "Combined: Downstream: Agnostic ! Upstream: Agnostic\n",
      "Upstream: Aware:kazachstania africana\n",
      "Combined: Downstream: Aware:candida glabrata ! Upstream: Aware:kazachstania africana\n"
     ]
    }
   ],
   "source": [
    "rows = []\n",
    "df_list = []\n",
    "total_len = y_obs.shape[0]\n",
    "test_indices = np.array(expr_merged[\"is_test\"])#.astype(int)\n",
    "np.random.seed(42)\n",
    "\n",
    "trained_models = {}\n",
    "\n",
    "for key in data_matrices:\n",
    "    print(key)\n",
    "    y = y_obs\n",
    "    X = data_matrices[key]\n",
    "    pipe = pipeline.make_pipeline(sklearn.preprocessing.StandardScaler(), sklearn.linear_model.RidgeCV(cv=5,alphas=[10,100,1000])) \n",
    "    pipe.fit(X[~test_indices],y[~test_indices])\n",
    "    trained_models[key] = pipe\n",
    "    y_pred = pipe.predict(X[test_indices])\n",
    "    r2_val = pearson_r2_metric(y_obs[test_indices], y_pred)\n",
    "    rows.append({\"model\":key, \"r2\":r2_val})\n",
    "    df_list.append(pd.DataFrame({\"model\":key,\"y_pred\":y_pred,\"y_true\":y_obs[test_indices]}))\n",
    "        \n",
    "metrics_expr = pd.DataFrame(rows)\n",
    "preds_expr = pd.concat(df_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "7335b3f8-f222-460c-833b-2765d08deaf5",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>model</th>\n",
       "      <th>r2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>3-mer-fiveprime</td>\n",
       "      <td>0.055139</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>4-mer-fiveprime</td>\n",
       "      <td>0.075567</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>5-mer-fiveprime</td>\n",
       "      <td>0.079634</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3-mer-threeprime</td>\n",
       "      <td>0.045927</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4-mer-threeprime</td>\n",
       "      <td>0.040213</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5-mer-threeprime</td>\n",
       "      <td>0.016925</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>3-mer-combined</td>\n",
       "      <td>0.078568</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>4-mer-combined</td>\n",
       "      <td>0.094468</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>5-mer-combined</td>\n",
       "      <td>0.073245</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Downstream: Agnostic</td>\n",
       "      <td>0.240178</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Downstream: Aware:candida glabrata</td>\n",
       "      <td>0.296830</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>Upstream: Agnostic</td>\n",
       "      <td>0.365009</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>Combined: Downstream: Agnostic ! Upstream: Agn...</td>\n",
       "      <td>0.437758</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>Upstream: Aware:kazachstania africana</td>\n",
       "      <td>0.491662</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>Combined: Downstream: Aware:candida glabrata !...</td>\n",
       "      <td>0.549527</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                model        r2\n",
       "0                                     3-mer-fiveprime  0.055139\n",
       "1                                     4-mer-fiveprime  0.075567\n",
       "2                                     5-mer-fiveprime  0.079634\n",
       "3                                    3-mer-threeprime  0.045927\n",
       "4                                    4-mer-threeprime  0.040213\n",
       "5                                    5-mer-threeprime  0.016925\n",
       "6                                      3-mer-combined  0.078568\n",
       "7                                      4-mer-combined  0.094468\n",
       "8                                      5-mer-combined  0.073245\n",
       "9                                Downstream: Agnostic  0.240178\n",
       "10                 Downstream: Aware:candida glabrata  0.296830\n",
       "11                                 Upstream: Agnostic  0.365009\n",
       "12  Combined: Downstream: Agnostic ! Upstream: Agn...  0.437758\n",
       "13              Upstream: Aware:kazachstania africana  0.491662\n",
       "14  Combined: Downstream: Aware:candida glabrata !...  0.549527"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics_expr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "31681b3e-de65-4c37-8aa5-7a99eb58b97b",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "tmp_data_matrices= dict()\n",
    "for key in data_matrices.keys():\n",
    "    tmp_data_matrices[key] = (data_matrices[key],trained_models[key][1].alpha_)\n",
    "data_matrices = tmp_data_matrices"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "id": "785206a0-ac67-47d5-b5c4-9345668235f3",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3-mer-fiveprime\n",
      "4-mer-fiveprime\n",
      "5-mer-fiveprime\n",
      "3-mer-threeprime\n",
      "4-mer-threeprime\n",
      "5-mer-threeprime\n",
      "3-mer-combined\n",
      "4-mer-combined\n",
      "5-mer-combined\n",
      "Downstream: Agnostic\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 69%|██████▉   | 69/100 [00:03<00:01, 20.35it/s]\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[90], line 21\u001b[0m\n\u001b[1;32m     19\u001b[0m smpl_indices \u001b[38;5;241m=\u001b[39m np\u001b[38;5;241m.\u001b[39mrandom\u001b[38;5;241m.\u001b[39mchoice(train_indices, size\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mlen\u001b[39m(train_indices))\n\u001b[1;32m     20\u001b[0m pipe \u001b[38;5;241m=\u001b[39m pipeline\u001b[38;5;241m.\u001b[39mmake_pipeline(sklearn\u001b[38;5;241m.\u001b[39mpreprocessing\u001b[38;5;241m.\u001b[39mStandardScaler(), sklearn\u001b[38;5;241m.\u001b[39mlinear_model\u001b[38;5;241m.\u001b[39mRidge(alpha\u001b[38;5;241m=\u001b[39malpha))\n\u001b[0;32m---> 21\u001b[0m \u001b[43mpipe\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfit\u001b[49m\u001b[43m(\u001b[49m\u001b[43mX\u001b[49m\u001b[43m[\u001b[49m\u001b[43msmpl_indices\u001b[49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\u001b[43my\u001b[49m\u001b[43m[\u001b[49m\u001b[43msmpl_indices\u001b[49m\u001b[43m]\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m     22\u001b[0m y_pred \u001b[38;5;241m=\u001b[39m pipe\u001b[38;5;241m.\u001b[39mpredict(X[test_indices])\n\u001b[1;32m     23\u001b[0m r2_val \u001b[38;5;241m=\u001b[39m pearson_r2_metric(y_obs[test_indices], y_pred)\n",
      "File \u001b[0;32m/opt/modules/i12g/anaconda/envs/motifnew/lib/python3.9/site-packages/sklearn/pipeline.py:405\u001b[0m, in \u001b[0;36mPipeline.fit\u001b[0;34m(self, X, y, **fit_params)\u001b[0m\n\u001b[1;32m    403\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_final_estimator \u001b[38;5;241m!=\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mpassthrough\u001b[39m\u001b[38;5;124m\"\u001b[39m:\n\u001b[1;32m    404\u001b[0m         fit_params_last_step \u001b[38;5;241m=\u001b[39m fit_params_steps[\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39msteps[\u001b[38;5;241m-\u001b[39m\u001b[38;5;241m1\u001b[39m][\u001b[38;5;241m0\u001b[39m]]\n\u001b[0;32m--> 405\u001b[0m         \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_final_estimator\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfit\u001b[49m\u001b[43m(\u001b[49m\u001b[43mXt\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43my\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mfit_params_last_step\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    407\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\n",
      "File \u001b[0;32m/opt/modules/i12g/anaconda/envs/motifnew/lib/python3.9/site-packages/sklearn/linear_model/_ridge.py:1134\u001b[0m, in \u001b[0;36mRidge.fit\u001b[0;34m(self, X, y, sample_weight)\u001b[0m\n\u001b[1;32m   1125\u001b[0m _accept_sparse \u001b[38;5;241m=\u001b[39m _get_valid_accept_sparse(sparse\u001b[38;5;241m.\u001b[39missparse(X), \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39msolver)\n\u001b[1;32m   1126\u001b[0m X, y \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_validate_data(\n\u001b[1;32m   1127\u001b[0m     X,\n\u001b[1;32m   1128\u001b[0m     y,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m   1132\u001b[0m     y_numeric\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mTrue\u001b[39;00m,\n\u001b[1;32m   1133\u001b[0m )\n\u001b[0;32m-> 1134\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfit\u001b[49m\u001b[43m(\u001b[49m\u001b[43mX\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43my\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43msample_weight\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43msample_weight\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m/opt/modules/i12g/anaconda/envs/motifnew/lib/python3.9/site-packages/sklearn/linear_model/_ridge.py:866\u001b[0m, in \u001b[0;36m_BaseRidge.fit\u001b[0;34m(self, X, y, sample_weight)\u001b[0m\n\u001b[1;32m    863\u001b[0m     sample_weight \u001b[38;5;241m=\u001b[39m _check_sample_weight(sample_weight, X, dtype\u001b[38;5;241m=\u001b[39mX\u001b[38;5;241m.\u001b[39mdtype)\n\u001b[1;32m    865\u001b[0m \u001b[38;5;66;03m# when X is sparse we only remove offset from y\u001b[39;00m\n\u001b[0;32m--> 866\u001b[0m X, y, X_offset, y_offset, X_scale \u001b[38;5;241m=\u001b[39m \u001b[43m_preprocess_data\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    867\u001b[0m \u001b[43m    \u001b[49m\u001b[43mX\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    868\u001b[0m \u001b[43m    \u001b[49m\u001b[43my\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    869\u001b[0m \u001b[43m    \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfit_intercept\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    870\u001b[0m \u001b[43m    \u001b[49m\u001b[43mcopy\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mcopy_X\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    871\u001b[0m \u001b[43m    \u001b[49m\u001b[43msample_weight\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43msample_weight\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    872\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    874\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m solver \u001b[38;5;241m==\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124msag\u001b[39m\u001b[38;5;124m\"\u001b[39m \u001b[38;5;129;01mand\u001b[39;00m sparse\u001b[38;5;241m.\u001b[39missparse(X) \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mfit_intercept:\n\u001b[1;32m    875\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcoef_, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mn_iter_, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mintercept_ \u001b[38;5;241m=\u001b[39m _ridge_regression(\n\u001b[1;32m    876\u001b[0m         X,\n\u001b[1;32m    877\u001b[0m         y,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    887\u001b[0m         check_input\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mFalse\u001b[39;00m,\n\u001b[1;32m    888\u001b[0m     )\n",
      "File \u001b[0;32m/opt/modules/i12g/anaconda/envs/motifnew/lib/python3.9/site-packages/sklearn/linear_model/_base.py:255\u001b[0m, in \u001b[0;36m_preprocess_data\u001b[0;34m(X, y, fit_intercept, normalize, copy, sample_weight, check_input)\u001b[0m\n\u001b[1;32m    252\u001b[0m         X_offset \u001b[38;5;241m=\u001b[39m np\u001b[38;5;241m.\u001b[39maverage(X, axis\u001b[38;5;241m=\u001b[39m\u001b[38;5;241m0\u001b[39m, weights\u001b[38;5;241m=\u001b[39msample_weight)\n\u001b[1;32m    254\u001b[0m     X_offset \u001b[38;5;241m=\u001b[39m X_offset\u001b[38;5;241m.\u001b[39mastype(X\u001b[38;5;241m.\u001b[39mdtype, copy\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mFalse\u001b[39;00m)\n\u001b[0;32m--> 255\u001b[0m     X \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m=\u001b[39m X_offset\n\u001b[1;32m    257\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m normalize:\n\u001b[1;32m    258\u001b[0m     X_var \u001b[38;5;241m=\u001b[39m X_var\u001b[38;5;241m.\u001b[39mastype(X\u001b[38;5;241m.\u001b[39mdtype, copy\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mFalse\u001b[39;00m)\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "#The boots trap\n",
    "\n",
    "rows = []\n",
    "df_list = []\n",
    "total_len = y_obs.shape[0]\n",
    "test_indices = np.array(expr_merged[\"is_test\"])#.astype(int)\n",
    "np.random.seed(42)\n",
    "\n",
    "train_indices = np.arange(len(test_indices))[~test_indices]\n",
    "bootstrapped_models = collections.defaultdict(list)\n",
    "\n",
    "for key in data_matrices:\n",
    "    print(key)\n",
    "    if not \"Downstream\" in key or \"6-mer\" in key:\n",
    "        continue\n",
    "    y = y_obs\n",
    "    X, alpha = data_matrices[key]\n",
    "    for i in tqdm.tqdm(range(100)):\n",
    "        smpl_indices = np.random.choice(train_indices, size=len(train_indices))\n",
    "        pipe = pipeline.make_pipeline(sklearn.preprocessing.StandardScaler(), sklearn.linear_model.Ridge(alpha=alpha))\n",
    "        pipe.fit(X[smpl_indices],y[smpl_indices])\n",
    "        y_pred = pipe.predict(X[test_indices])\n",
    "        r2_val = pearson_r2_metric(y_obs[test_indices], y_pred)\n",
    "        rows.append({\"model\":key, \"r2\":r2_val})\n",
    "        bootstrapped_models[key].append(pipe)\n",
    "\n",
    "metrics_expr_bootstrap = pd.DataFrame(rows)\n",
    "\n",
    "metrics_expr_bootstrap.groupby('model').std() * 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "id": "04ac2e1b-9e18-4ee4-a073-e2d8b1ce2d28",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'metrics_expr_bootstrap' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[91], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m \u001b[43mmetrics_expr_bootstrap\u001b[49m\u001b[38;5;241m.\u001b[39mgroupby(\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mmodel\u001b[39m\u001b[38;5;124m'\u001b[39m)\u001b[38;5;241m.\u001b[39mstd()\u001b[38;5;241m.\u001b[39msort_values(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mr2\u001b[39m\u001b[38;5;124m\"\u001b[39m) \u001b[38;5;241m*\u001b[39m \u001b[38;5;241m2\u001b[39m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'metrics_expr_bootstrap' is not defined"
     ]
    }
   ],
   "source": [
    "metrics_expr_bootstrap.groupby('model').std().sort_values(\"r2\") * 2"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2470f342-b8a3-4d83-8801-459dfe8af742",
   "metadata": {},
   "source": [
    "# Yamanishi"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "id": "c17a5f65-91eb-4a34-99d5-e47323ad3b73",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def yamanishi_mpra(path_list, trained_models):\n",
    "    yamanishi = pd.read_csv(\"./data/Downstream_Targets/yamanishi_terminator.tsv\",sep=\"\\t\")#[\"Sequence of amplified region\"].str.len().describe()\n",
    "    zrimec_preds = pd.read_csv(\"./data/Downstream_Targets/df_Yamanishi2013_predictions.csv\")\n",
    "    yamanishi_emb_dict = {} \n",
    "    for path in path_list:\n",
    "        tmp = np.load(f\"{project_path}/{path}/embeddings.npy\")            \n",
    "        yamanishi_emb_dict[get_model_name(path)] =  np.array([tmp[i] for i in range(tmp.shape[0])])\n",
    "    for model_name, emb in yamanishi_emb_dict.items():\n",
    "        print (model_name, emb.shape)\n",
    "        yamanishi[model_name] = [emb[i] for i in range(emb.shape[0])]\n",
    "    yamanishi = yamanishi.merge(zrimec_preds, left_on = \"Systematic Name\", right_on = \"gene_name\")\n",
    "    y_obs = np.array(yamanishi[\"relative FI\"])\n",
    "    X_base = np.zeros((len(y_obs),768))    \n",
    "    X_models = {}\n",
    "    data_matrices = {}\n",
    "    for model in yamanishi_emb_dict.keys():\n",
    "        if 'Nucleotide' in model:\n",
    "            X_models[model] = np.concatenate([np.zeros((len(y_obs), 2560)), np.stack(yamanishi[model],axis=0)],axis=1)\n",
    "        else:\n",
    "            try:\n",
    "                X_models[model] = np.concatenate([X_base, np.stack(yamanishi[model],axis=0)],axis=1)\n",
    "\n",
    "            except:\n",
    "                print (\"could not stack base\",model)\n",
    "        try:\n",
    "            data_matrices[model] = X_models[model],trained_models[f\"Combined: Downstream: {model} ! Upstream: Agnostic\"]\n",
    "            continue\n",
    "        except:\n",
    "            pass\n",
    "        try:\n",
    "            data_matrices[model] = X_models[model],trained_models[f\"Combined: Downstream: {model} ! Upstream: Aware:kazachstania africana\"]\n",
    "        except:\n",
    "            print (\"No matching combined model for:\",model)\n",
    "    rows = []\n",
    "    dataset = yamanishi\n",
    "    seq_col = \"Sequence of amplified region\"\n",
    "    dataset[seq_col] = \"TAA\" + dataset[seq_col]\n",
    "    dataset[seq_col] = dataset[seq_col].str[:300] # 509        \n",
    "    X_3mer_three = np.stack(dataset[seq_col].apply(lambda x: kmerizer3.kmerize(x)))\n",
    "    X_4mer_three = np.stack(dataset[seq_col].apply(lambda x: kmerizer4.kmerize(x)))\n",
    "    X_5mer_three = np.stack(dataset[seq_col].apply(lambda x: kmerizer5.kmerize(x)))\n",
    "    #X_6mer_three = np.concatenate([X_base, np.stack(dataset[seq_col].apply(lambda x: kmerizer6.kmerize(x)))],axis=1)\n",
    "    X_3mer_combined = np.concatenate([np.zeros(X_3mer_three.shape), X_3mer_three],axis=1)\n",
    "    X_4mer_combined = np.concatenate([np.zeros(X_4mer_three.shape), X_4mer_three],axis=1)\n",
    "    X_5mer_combined = np.concatenate([np.zeros(X_5mer_three.shape), X_5mer_three],axis=1)\n",
    "    data_matrices[\"3-mer\" ] = X_3mer_combined,trained_models['3-mer-combined']\n",
    "    data_matrices[\"4-mer\" ] = X_4mer_combined,trained_models['4-mer-combined']\n",
    "    data_matrices[\"5-mer\" ] = X_5mer_combined,trained_models['5-mer-combined']\n",
    "\n",
    "    for key in data_matrices:\n",
    "        print (key)\n",
    "        X = data_matrices[key][0]\n",
    "        print (X.shape)\n",
    "        pipe = data_matrices[key][1]\n",
    "        y_pred = pipe.predict(X)\n",
    "        r2_val = pearson_r2_metric(y_obs, y_pred)\n",
    "        rows.append({\"model\":key, \"r2\":r2_val})    \n",
    "    rows.append({\"model\":\"zrimec\",\"r2\":pearson_r2_metric(y_obs, yamanishi[\"ypred_tdh3\"])})\n",
    "    return pd.DataFrame(rows)\n",
    "\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "id": "7667b89c-9ef5-4f92-94b9-44731d9d172c",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Agnostic embedding-downstream_300-static_agnostic-yamanishi-512_withstop_bertstyle\n",
      "Aware:candida glabrata embedding-species_downstream_300-static_candida_glabrata-yamanishi-512_withstop_bertstyle\n",
      "Agnostic (5302, 768)\n",
      "Aware:candida glabrata (5302, 768)\n",
      "Agnostic\n",
      "(4005, 1536)\n",
      "Aware:candida glabrata\n",
      "(4005, 1536)\n",
      "3-mer\n",
      "(4005, 128)\n",
      "4-mer\n",
      "(4005, 512)\n",
      "5-mer\n",
      "(4005, 2048)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>model</th>\n",
       "      <th>r2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Agnostic</td>\n",
       "      <td>0.070537</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Aware:candida glabrata</td>\n",
       "      <td>0.107829</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3-mer</td>\n",
       "      <td>0.019974</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4-mer</td>\n",
       "      <td>0.029193</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5-mer</td>\n",
       "      <td>0.024310</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>zrimec</td>\n",
       "      <td>0.099132</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                    model        r2\n",
       "0                Agnostic  0.070537\n",
       "1  Aware:candida glabrata  0.107829\n",
       "2                   3-mer  0.019974\n",
       "3                   4-mer  0.029193\n",
       "4                   5-mer  0.024310\n",
       "5                  zrimec  0.099132"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics_yamanishi = yamanishi_mpra(dataset_and_model_map[\"yamanishi\"], trained_models)\n",
    "metrics_yamanishi"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3a3f1959-6427-4095-8675-cedbd611279c",
   "metadata": {
    "tags": []
   },
   "source": [
    "# Segal Promoter MPRA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "id": "fd9b00d8-0535-4ce1-bb66-549aad7e8055",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def segal_promoter_mpra(path_list, trained_models):\n",
    "    test_predictions = []\n",
    "    segal_seq = pd.read_csv(\"./data/Downstream_Targets/segal_sequences.tsv\",sep=\"\\t\")\n",
    "    segal_expr = pd.read_csv(\"./data/Downstream_Targets/segal_expression.tsv\",sep=\"\\t\")\n",
    "    zrimec_preds = pd.read_csv(\"./data/Downstream_Targets/df_Keren2013_predictions.csv\")\n",
    "    # five prime, no species\n",
    "    segal_emb_dict = dict()\n",
    "    for path in path_list:\n",
    "        tmp = np.load(f\"{project_path}/{path}/embeddings.npy\")  \n",
    "        segal_emb_dict[get_model_name(path)] = [tmp[i] for i in range(tmp.shape[0])]\n",
    "        \n",
    "    for model_name, emb in segal_emb_dict.items():\n",
    "        segal_seq[model_name] = emb\n",
    "    segal_5prime_df = segal_seq.merge(segal_expr, on=\"ORF\")\n",
    "    segal_5prime_df = segal_5prime_df.query('Seq != \"NoSeq\"')\n",
    "    segal_5prime_df = segal_5prime_df.loc[~segal_5prime_df.Seq.str.contains(\":\")]\n",
    "    \n",
    "    segal_5prime_zrimec = segal_5prime_df.merge(zrimec_preds[[\"ORF\",\"ypred\"]],on=\"ORF\")\n",
    "    y_obs = np.array(segal_5prime_zrimec[['Glucose', 'Fructose', 'Sorbitol 1M', 'Growth at 39o',\n",
    "       'NaCl 1M', 'Glucose lacking amino acids',\n",
    "       'Galactose lacking amino acids', 'Galactose', 'Glycerol', 'Ethanol']].mean(axis=1))\n",
    "    X_models = {}\n",
    "    for model_name in segal_emb_dict.keys():    \n",
    "        X_models[model_name] = np.stack(segal_5prime_zrimec[model_name],axis=0)\n",
    "        X_models[model_name] = np.concatenate([X_models[model_name], np.zeros(X_models[model_name].shape)],axis=1)\n",
    "    data_matrices = {}\n",
    "    for model in X_models.keys():\n",
    "        try:\n",
    "            data_matrices[f\"Combined: Downstream: {model} ! Upstream: Agnostic\"] = X_models[model],trained_models[f\"Combined: Downstream: {model} ! Upstream: Agnostic\"]\n",
    "            continue\n",
    "        except:\n",
    "            pass\n",
    "        try:\n",
    "            data_matrices[f\"Combined: Downstream: Aware:candida glabrata ! Upstream: {model}\"] = X_models[model],trained_models[f\"Combined: Downstream: Aware:candida glabrata ! Upstream: {model}\"]\n",
    "        except:\n",
    "            print (\"No matching combined model for:\",model)\n",
    "\n",
    "    dataset = segal_5prime_zrimec\n",
    "    #return dataset, y_obs\n",
    "    seq_col = \"Seq\"\n",
    "    X_3mer_five = np.stack(dataset[seq_col].apply(lambda x: kmerizer3.kmerize(x)))\n",
    "    X_4mer_five = np.stack(dataset[seq_col].apply(lambda x: kmerizer4.kmerize(x)))\n",
    "    X_5mer_five = np.stack(dataset[seq_col].apply(lambda x: kmerizer5.kmerize(x)))\n",
    "    #X_6mer_five = np.concatenate([X_base, np.stack(dataset[seq_col].apply(lambda x: kmerizer6.kmerize(x)))],axis=1)\n",
    "    print(X_3mer_five.shape)\n",
    "    X_3mer_combined = np.concatenate([X_3mer_five,np.zeros(X_3mer_five.shape)],axis=1)\n",
    "    X_4mer_combined = np.concatenate([X_4mer_five, np.zeros(X_4mer_five.shape)],axis=1)\n",
    "    X_5mer_combined = np.concatenate([X_5mer_five,np.zeros(X_5mer_five.shape)],axis=1)\n",
    "    data_matrices[\"3-mer\" ] = X_3mer_combined,trained_models['3-mer-combined']\n",
    "    data_matrices[\"4-mer\" ] = X_4mer_combined,trained_models['4-mer-combined']\n",
    "    data_matrices[\"5-mer\" ] = X_5mer_combined,trained_models['5-mer-combined']\n",
    "    \n",
    "    \n",
    "    rows = []\n",
    "    for key in data_matrices:\n",
    "        X = data_matrices[key][0]\n",
    "        pipe = data_matrices[key][1]\n",
    "        y_pred = pipe.predict(X)\n",
    "        r2_val = pearson_r2_metric(y_obs, y_pred)\n",
    "        rows.append({\"model\":key, \"r2\":r2_val})\n",
    "        test_predictions.append(pd.DataFrame({\"model\":key,\n",
    "                                 \"pred\": y_pred, \n",
    "                                 \"true\": y_obs}))\n",
    "    \n",
    "    rows.append({\"model\":\"zrimec\",\"r2\":pearson_r2_metric(y_obs, segal_5prime_zrimec[\"ypred\"])})\n",
    "    test_predictions.append(pd.DataFrame({\"model\":'zrimec',\n",
    "                                 \"pred\": segal_5prime_zrimec[\"ypred\"], \n",
    "                                 \"true\": y_obs}))\n",
    "    pred_micro_scer = pd.concat(test_predictions)\n",
    "    pred_micro_scer[\"residual\"] = (pred_micro_scer[\"true\"] - pred_micro_scer[\"pred\"])**2\n",
    "    return pd.DataFrame(rows),pred_micro_scer\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "id": "d7f6716d-d625-409e-86a5-ce7ab4598247",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Aware:kazachstania africana embedding-species_upstream_1000-static_kazachstania_africana_cbs_2517_gca_000304475-segal_5prime-1024_withoutstop_bertstyle\n",
      "Agnostic embedding-upstream_1000-static_agnostic-segal_5prime-1024_withoutstop_bertstyle\n",
      "(621, 64)\n"
     ]
    }
   ],
   "source": [
    "segal_promoter_mpra_metrics, pred_micro_scer = segal_promoter_mpra(dataset_and_model_map[\"segal_5prime\"], trained_models)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "id": "488e7463-9014-45da-bcb3-73e562f2c4e4",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>model</th>\n",
       "      <th>r2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Combined: Downstream: Aware:candida glabrata !...</td>\n",
       "      <td>0.453971</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Combined: Downstream: Agnostic ! Upstream: Agn...</td>\n",
       "      <td>0.351996</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3-mer</td>\n",
       "      <td>0.006833</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4-mer</td>\n",
       "      <td>0.061807</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5-mer</td>\n",
       "      <td>0.230468</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>zrimec</td>\n",
       "      <td>0.440276</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                               model        r2\n",
       "0  Combined: Downstream: Aware:candida glabrata !...  0.453971\n",
       "1  Combined: Downstream: Agnostic ! Upstream: Agn...  0.351996\n",
       "2                                              3-mer  0.006833\n",
       "3                                              4-mer  0.061807\n",
       "4                                              5-mer  0.230468\n",
       "5                                             zrimec  0.440276"
      ]
     },
     "execution_count": 102,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "segal_promoter_mpra_metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "id": "ab6f9940-fe14-4f47-ab06-2124a915110f",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "WilcoxonResult(statistic=0.0, pvalue=2.233932560757222e-103)"
      ]
     },
     "execution_count": 103,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scipy.stats.wilcoxon(pred_micro_scer.query('model == \"Combined: Downstream: Aware:candida glabrata ! Upstream: Aware:kazachstania africana\"')[\"residual\"], pred_micro_scer.query('model == \"zrimec\"')[\"residual\"])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "32e6180d-c6b2-4baa-a073-7d613a693430",
   "metadata": {},
   "source": [
    "# Predict for diauxic shift Microarray data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "ce1f1a74-7bdc-4ff8-b11f-4eb92081648b",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "microarray_all = pd.read_csv(\"./data/Downstream_Targets/Microarray_raw_dedup.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "8f20e5b6-2f6d-4b29-b1a4-a2a2b3916f39",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "organisms = ['candida_albicans_sc5314_gca_000784635', 'candida_glabrata', 'debaryomyces_hansenii_cbs767_gca_000006445',\n",
    " 'kluyveromyces_lactis_gca_000002515', 'schizosaccharomyces_japonicus', \"schizosaccharomyces_pombe\",\n",
    " 'vanderwaltozyma_polyspora_dsm_70294_gca_000150035', 'yarrowia_lipolytica', 'saccharomyces_cerevisiae']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "af885580-e9bf-4684-b787-6ee721cb3a9c",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "five_sequences_all = pd.read_parquet(\"./data/Sequences/Annotation/Assembled/five_prime.parquet\")\n",
    "five_sequences_all = five_sequences_all.query('organism in @organisms')\n",
    "five_sequences_all = five_sequences_all.loc[five_sequences_all[\"five_prime_seq\"].str.len() == 1003]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "26f14ec9-354c-4ebe-aed9-6cf468b8510d",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "emb_fiveprime = np.load(project_path+\"embedding-upstream_1000-static_agnostic-microarray-1024_withoutstop_bertstyle/embeddings.npy\")\n",
    "emb_fiveprime = [emb_fiveprime[i] for i in range(emb_fiveprime.shape[0])]\n",
    "five_sequences_all[\"five_prime_embedding\"] = emb_fiveprime\n",
    "\n",
    "emb_fiveprime_species = np.load(project_path+\"embedding-species_upstream_1000-static_many_species-microarray-1024_withoutstop_bertstyle/embeddings.npy\")\n",
    "emb_fiveprime_species = [emb_fiveprime_species[i] for i in range(emb_fiveprime_species.shape[0])]\n",
    "five_sequences_all[\"five_prime_embedding_species\"] = emb_fiveprime_species\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "5d68f01a-1b47-42bb-8efe-884b2a08580d",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "organisms_threeprime = ['candida_albicans_sc5314_gca_000784635', 'candida_glabrata', 'debaryomyces_hansenii_cbs767_gca_000006445',\n",
    "                 'kluyveromyces_lactis_gca_000002515',\n",
    "                 'vanderwaltozyma_polyspora_dsm_70294_gca_000150035', 'yarrowia_lipolytica']\n",
    "\n",
    "three_sequences_all = pd.read_parquet(\"./data/Sequences/Annotation/Assembled/three_prime.parquet\")\n",
    "three_sequences_all = three_sequences_all.query('organism in @organisms_threeprime')\n",
    "three_sequences_all[\"three_prime_seq\"] = three_sequences_all[\"three_prime_seq\"].str[:300]\n",
    "three_sequences_all = three_sequences_all.loc[three_sequences_all[\"three_prime_seq\"].str.len() == 300]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "6be3fb69-b227-4a59-91d4-ae4fc69f94f7",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "emb_threeprime = np.load(project_path+\"embedding-downstream_300-static_agnostic-microarray_downstream-512_withstop_bertstyle/embeddings.npy\")\n",
    "emb_threeprime = [emb_threeprime[i] for i in range(emb_threeprime.shape[0])]\n",
    "three_sequences_all[\"three_prime_embedding\"] = emb_threeprime\n",
    "\n",
    "emb_threeprime_species = np.load(project_path+\"embedding-species_downstream_300-static_many_species-microarray_downstream-512_withstop_bertstyle/embeddings.npy\")\n",
    "emb_threeprime_species = [emb_threeprime_species[i] for i in range(emb_threeprime_species.shape[0])]\n",
    "three_sequences_all[\"three_prime_embedding_species\"] = emb_threeprime_species"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "82f6b441-7f60-44cd-bfef-977756569596",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>gene_id</th>\n",
       "      <th>transcript_id</th>\n",
       "      <th>Chromosome</th>\n",
       "      <th>Feature</th>\n",
       "      <th>Start</th>\n",
       "      <th>End</th>\n",
       "      <th>Strand</th>\n",
       "      <th>five_prime_start</th>\n",
       "      <th>five_prime_end</th>\n",
       "      <th>five_prime_seq</th>\n",
       "      <th>organism</th>\n",
       "      <th>species</th>\n",
       "      <th>five_prime_embedding</th>\n",
       "      <th>five_prime_embedding_species</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3435471</th>\n",
       "      <td>W5Q_00575</td>\n",
       "      <td>KHC83359</td>\n",
       "      <td>supercont4.1</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>1234961</td>\n",
       "      <td>1234964</td>\n",
       "      <td>+</td>\n",
       "      <td>1233961</td>\n",
       "      <td>1234964</td>\n",
       "      <td>CCTCGTTTGAAATGGTAGTTTTGAAATAGGGTTTGTATCCAACAGG...</td>\n",
       "      <td>candida_albicans_sc5314_gca_000784635</td>\n",
       "      <td>candida_albicans</td>\n",
       "      <td>[0.068653494, 0.015206551, -0.021652974, -0.16...</td>\n",
       "      <td>[0.06493876, -0.100349575, 0.018514436, 0.0325...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3435472</th>\n",
       "      <td>W5Q_00675</td>\n",
       "      <td>KHC83458</td>\n",
       "      <td>supercont4.1</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>1488618</td>\n",
       "      <td>1488621</td>\n",
       "      <td>+</td>\n",
       "      <td>1487618</td>\n",
       "      <td>1488621</td>\n",
       "      <td>ACCAGTATGTAAGGAGATGATGCTAACAGAGGATTCAGTTTTGGTG...</td>\n",
       "      <td>candida_albicans_sc5314_gca_000784635</td>\n",
       "      <td>candida_albicans</td>\n",
       "      <td>[0.036147296, 0.025376495, -0.0061738696, -0.0...</td>\n",
       "      <td>[0.0545368, 0.119643144, -0.0065993257, 0.0437...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3435473</th>\n",
       "      <td>W5Q_00708</td>\n",
       "      <td>KHC83490</td>\n",
       "      <td>supercont4.1</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>1564846</td>\n",
       "      <td>1564849</td>\n",
       "      <td>+</td>\n",
       "      <td>1563846</td>\n",
       "      <td>1564849</td>\n",
       "      <td>AATTTTGACATTCAATATACCAGGACTGACTATACAATTGATTTCA...</td>\n",
       "      <td>candida_albicans_sc5314_gca_000784635</td>\n",
       "      <td>candida_albicans</td>\n",
       "      <td>[-0.15883088, -0.021592634, 0.030859975, -0.54...</td>\n",
       "      <td>[0.01612491, 0.086933665, -0.031225346, 0.0250...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3435474</th>\n",
       "      <td>W5Q_01237</td>\n",
       "      <td>KHC84013</td>\n",
       "      <td>supercont4.1</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>2807108</td>\n",
       "      <td>2807111</td>\n",
       "      <td>+</td>\n",
       "      <td>2806108</td>\n",
       "      <td>2807111</td>\n",
       "      <td>ACAATTATTTGGTGATGATAATGGAGAATTTAATTTAAGTGAAATA...</td>\n",
       "      <td>candida_albicans_sc5314_gca_000784635</td>\n",
       "      <td>candida_albicans</td>\n",
       "      <td>[-0.089624725, 0.02925052, -0.014624437, -0.90...</td>\n",
       "      <td>[0.077558234, 0.3897506, -0.015687384, 0.03565...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3435475</th>\n",
       "      <td>W5Q_00536</td>\n",
       "      <td>KHC83323</td>\n",
       "      <td>supercont4.1</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>1156335</td>\n",
       "      <td>1156338</td>\n",
       "      <td>+</td>\n",
       "      <td>1155335</td>\n",
       "      <td>1156338</td>\n",
       "      <td>AATGATTTAGTACCATTTTCGATACTCTTTTTCCCACTTTCAGTAG...</td>\n",
       "      <td>candida_albicans_sc5314_gca_000784635</td>\n",
       "      <td>candida_albicans</td>\n",
       "      <td>[-0.073895775, 0.055318058, -0.003053937, -0.7...</td>\n",
       "      <td>[0.02883135, -0.09521633, -0.027411956, 0.0001...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14677878</th>\n",
       "      <td>YALI0_F18040g</td>\n",
       "      <td>CAG78370</td>\n",
       "      <td>F</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>2404344</td>\n",
       "      <td>2404347</td>\n",
       "      <td>-</td>\n",
       "      <td>2404344</td>\n",
       "      <td>2405347</td>\n",
       "      <td>TATTGTAGTGGGAGACCATACGAGAATCCTGGGTGCTGTAATCTTT...</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>[-0.3018363, -0.005869295, 0.022159437, -0.244...</td>\n",
       "      <td>[0.07424399, 0.12904361, -0.010953401, 0.04557...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14677881</th>\n",
       "      <td>YALI0_F10923g</td>\n",
       "      <td>CAG78073</td>\n",
       "      <td>F</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>1458078</td>\n",
       "      <td>1458081</td>\n",
       "      <td>-</td>\n",
       "      <td>1458078</td>\n",
       "      <td>1459081</td>\n",
       "      <td>GCCTCTTTCTCGAGTCGATCCGTCCACTGCAACCTCCATGGTCCCT...</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>[-0.46078548, -0.08309572, 0.038748596, -0.662...</td>\n",
       "      <td>[0.0104980245, 0.2365075, -0.0038241881, 0.047...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14677882</th>\n",
       "      <td>YALI0_F06556g</td>\n",
       "      <td>CAG77892</td>\n",
       "      <td>F</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>971905</td>\n",
       "      <td>971908</td>\n",
       "      <td>-</td>\n",
       "      <td>971905</td>\n",
       "      <td>972908</td>\n",
       "      <td>AATCAGACAATAGTACAAGTACAATGGAAGCTGGCCATTGTTGTCT...</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>[-0.37120572, -0.0012745588, 0.04519816, -0.07...</td>\n",
       "      <td>[0.10203641, 0.16689563, 0.0025999122, 0.00292...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14677883</th>\n",
       "      <td>YALI0_F26455g</td>\n",
       "      <td>CAG78720</td>\n",
       "      <td>F</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>3387003</td>\n",
       "      <td>3387006</td>\n",
       "      <td>-</td>\n",
       "      <td>3387003</td>\n",
       "      <td>3388006</td>\n",
       "      <td>ATACGGATCCCGCAATCTCTGGTCATAGTAGGAATGTCGAACGTAC...</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>[-0.18022203, -0.038665965, 0.030178862, -0.29...</td>\n",
       "      <td>[0.050544992, 0.04968849, -0.0055678324, 0.034...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14677884</th>\n",
       "      <td>YALI0_F28061g</td>\n",
       "      <td>CAG78787</td>\n",
       "      <td>F</td>\n",
       "      <td>start_codon</td>\n",
       "      <td>3575467</td>\n",
       "      <td>3575470</td>\n",
       "      <td>-</td>\n",
       "      <td>3575467</td>\n",
       "      <td>3576470</td>\n",
       "      <td>ATCTTGTTGTTCTCCTCCAATAACGACATGGGCAGCTTGTTTTGTC...</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>yarrowia_lipolytica</td>\n",
       "      <td>[-0.22522137, -0.016432116, 0.038102612, -0.51...</td>\n",
       "      <td>[0.07778198, 0.111588046, 0.0021588213, 0.0858...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>50973 rows × 14 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                gene_id transcript_id    Chromosome      Feature    Start   \n",
       "3435471       W5Q_00575      KHC83359  supercont4.1  start_codon  1234961  \\\n",
       "3435472       W5Q_00675      KHC83458  supercont4.1  start_codon  1488618   \n",
       "3435473       W5Q_00708      KHC83490  supercont4.1  start_codon  1564846   \n",
       "3435474       W5Q_01237      KHC84013  supercont4.1  start_codon  2807108   \n",
       "3435475       W5Q_00536      KHC83323  supercont4.1  start_codon  1156335   \n",
       "...                 ...           ...           ...          ...      ...   \n",
       "14677878  YALI0_F18040g      CAG78370             F  start_codon  2404344   \n",
       "14677881  YALI0_F10923g      CAG78073             F  start_codon  1458078   \n",
       "14677882  YALI0_F06556g      CAG77892             F  start_codon   971905   \n",
       "14677883  YALI0_F26455g      CAG78720             F  start_codon  3387003   \n",
       "14677884  YALI0_F28061g      CAG78787             F  start_codon  3575467   \n",
       "\n",
       "              End Strand  five_prime_start  five_prime_end   \n",
       "3435471   1234964      +           1233961         1234964  \\\n",
       "3435472   1488621      +           1487618         1488621   \n",
       "3435473   1564849      +           1563846         1564849   \n",
       "3435474   2807111      +           2806108         2807111   \n",
       "3435475   1156338      +           1155335         1156338   \n",
       "...           ...    ...               ...             ...   \n",
       "14677878  2404347      -           2404344         2405347   \n",
       "14677881  1458081      -           1458078         1459081   \n",
       "14677882   971908      -            971905          972908   \n",
       "14677883  3387006      -           3387003         3388006   \n",
       "14677884  3575470      -           3575467         3576470   \n",
       "\n",
       "                                             five_prime_seq   \n",
       "3435471   CCTCGTTTGAAATGGTAGTTTTGAAATAGGGTTTGTATCCAACAGG...  \\\n",
       "3435472   ACCAGTATGTAAGGAGATGATGCTAACAGAGGATTCAGTTTTGGTG...   \n",
       "3435473   AATTTTGACATTCAATATACCAGGACTGACTATACAATTGATTTCA...   \n",
       "3435474   ACAATTATTTGGTGATGATAATGGAGAATTTAATTTAAGTGAAATA...   \n",
       "3435475   AATGATTTAGTACCATTTTCGATACTCTTTTTCCCACTTTCAGTAG...   \n",
       "...                                                     ...   \n",
       "14677878  TATTGTAGTGGGAGACCATACGAGAATCCTGGGTGCTGTAATCTTT...   \n",
       "14677881  GCCTCTTTCTCGAGTCGATCCGTCCACTGCAACCTCCATGGTCCCT...   \n",
       "14677882  AATCAGACAATAGTACAAGTACAATGGAAGCTGGCCATTGTTGTCT...   \n",
       "14677883  ATACGGATCCCGCAATCTCTGGTCATAGTAGGAATGTCGAACGTAC...   \n",
       "14677884  ATCTTGTTGTTCTCCTCCAATAACGACATGGGCAGCTTGTTTTGTC...   \n",
       "\n",
       "                                       organism              species   \n",
       "3435471   candida_albicans_sc5314_gca_000784635     candida_albicans  \\\n",
       "3435472   candida_albicans_sc5314_gca_000784635     candida_albicans   \n",
       "3435473   candida_albicans_sc5314_gca_000784635     candida_albicans   \n",
       "3435474   candida_albicans_sc5314_gca_000784635     candida_albicans   \n",
       "3435475   candida_albicans_sc5314_gca_000784635     candida_albicans   \n",
       "...                                         ...                  ...   \n",
       "14677878                    yarrowia_lipolytica  yarrowia_lipolytica   \n",
       "14677881                    yarrowia_lipolytica  yarrowia_lipolytica   \n",
       "14677882                    yarrowia_lipolytica  yarrowia_lipolytica   \n",
       "14677883                    yarrowia_lipolytica  yarrowia_lipolytica   \n",
       "14677884                    yarrowia_lipolytica  yarrowia_lipolytica   \n",
       "\n",
       "                                       five_prime_embedding   \n",
       "3435471   [0.068653494, 0.015206551, -0.021652974, -0.16...  \\\n",
       "3435472   [0.036147296, 0.025376495, -0.0061738696, -0.0...   \n",
       "3435473   [-0.15883088, -0.021592634, 0.030859975, -0.54...   \n",
       "3435474   [-0.089624725, 0.02925052, -0.014624437, -0.90...   \n",
       "3435475   [-0.073895775, 0.055318058, -0.003053937, -0.7...   \n",
       "...                                                     ...   \n",
       "14677878  [-0.3018363, -0.005869295, 0.022159437, -0.244...   \n",
       "14677881  [-0.46078548, -0.08309572, 0.038748596, -0.662...   \n",
       "14677882  [-0.37120572, -0.0012745588, 0.04519816, -0.07...   \n",
       "14677883  [-0.18022203, -0.038665965, 0.030178862, -0.29...   \n",
       "14677884  [-0.22522137, -0.016432116, 0.038102612, -0.51...   \n",
       "\n",
       "                               five_prime_embedding_species  \n",
       "3435471   [0.06493876, -0.100349575, 0.018514436, 0.0325...  \n",
       "3435472   [0.0545368, 0.119643144, -0.0065993257, 0.0437...  \n",
       "3435473   [0.01612491, 0.086933665, -0.031225346, 0.0250...  \n",
       "3435474   [0.077558234, 0.3897506, -0.015687384, 0.03565...  \n",
       "3435475   [0.02883135, -0.09521633, -0.027411956, 0.0001...  \n",
       "...                                                     ...  \n",
       "14677878  [0.07424399, 0.12904361, -0.010953401, 0.04557...  \n",
       "14677881  [0.0104980245, 0.2365075, -0.0038241881, 0.047...  \n",
       "14677882  [0.10203641, 0.16689563, 0.0025999122, 0.00292...  \n",
       "14677883  [0.050544992, 0.04968849, -0.0055678324, 0.034...  \n",
       "14677884  [0.07778198, 0.111588046, 0.0021588213, 0.0858...  \n",
       "\n",
       "[50973 rows x 14 columns]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "five_sequences_all"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "67d49e81-dacd-44d1-bed2-1481df613e12",
   "metadata": {},
   "source": [
    "## C. glabrata\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "763d9929-906c-4478-b62a-7ed8371714d6",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "microarray_test = microarray_all.reset_index().query('Study == \"Thompson_diauxicshift\"').query('Organism not in [\"Saccharomyces cerevisiae\", \"Schizosaccharomyces pombe\", \"Schizosaccharomyces japonicus\"]')\n",
    "microarray_test = (microarray_test\n",
    "              .merge(five_sequences_all,on=\"gene_id\")\n",
    "              .merge(three_sequences_all[[\"gene_id\",\"three_prime_seq\",\"three_prime_embedding\",\"three_prime_embedding_species\"]],on=\"gene_id\")\n",
    "              .dropna())\n",
    "microarray = microarray_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "660f361f-00e5-4e51-b59f-d3d8b18c5988",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "microarray = microarray.query(\"Organism == 'Candida glabrata'\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "ee2ce997-bf93-43a8-9e07-b8e1e7336f46",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Candida glabrata\n",
      "PS\n",
      "BERT-ADN-threeprime\n",
      "BERT-ADN-threeprime-species\n",
      "BERT-ADN-fiveprime\n",
      "BERT-ADN-fiveprime-species\n",
      "BERT-ADN-combined\n",
      "BERT-ADN-combined-species\n",
      "3-mer-fiveprime\n",
      "4-mer-fiveprime\n",
      "5-mer-fiveprime\n",
      "3-mer-threeprime\n",
      "4-mer-threeprime\n",
      "5-mer-threeprime\n",
      "3-mer-combined\n",
      "4-mer-combined\n",
      "5-mer-combined\n",
      "PLAT\n",
      "BERT-ADN-threeprime\n",
      "BERT-ADN-threeprime-species\n",
      "BERT-ADN-fiveprime\n",
      "BERT-ADN-fiveprime-species\n",
      "BERT-ADN-combined\n",
      "BERT-ADN-combined-species\n",
      "3-mer-fiveprime\n",
      "4-mer-fiveprime\n",
      "5-mer-fiveprime\n",
      "3-mer-threeprime\n",
      "4-mer-threeprime\n",
      "5-mer-threeprime\n",
      "3-mer-combined\n",
      "4-mer-combined\n",
      "5-mer-combined\n",
      "LAG\n",
      "BERT-ADN-threeprime\n",
      "BERT-ADN-threeprime-species\n",
      "BERT-ADN-fiveprime\n",
      "BERT-ADN-fiveprime-species\n",
      "BERT-ADN-combined\n",
      "BERT-ADN-combined-species\n",
      "3-mer-fiveprime\n",
      "4-mer-fiveprime\n",
      "5-mer-fiveprime\n",
      "3-mer-threeprime\n",
      "4-mer-threeprime\n",
      "5-mer-threeprime\n",
      "3-mer-combined\n",
      "4-mer-combined\n",
      "5-mer-combined\n",
      "ll\n",
      "BERT-ADN-threeprime\n",
      "BERT-ADN-threeprime-species\n",
      "BERT-ADN-fiveprime\n",
      "BERT-ADN-fiveprime-species\n",
      "BERT-ADN-combined\n",
      "BERT-ADN-combined-species\n",
      "3-mer-fiveprime\n",
      "4-mer-fiveprime\n",
      "5-mer-fiveprime\n",
      "3-mer-threeprime\n",
      "4-mer-threeprime\n",
      "5-mer-threeprime\n",
      "3-mer-combined\n",
      "4-mer-combined\n",
      "5-mer-combined\n",
      "DS\n",
      "BERT-ADN-threeprime\n",
      "BERT-ADN-threeprime-species\n",
      "BERT-ADN-fiveprime\n",
      "BERT-ADN-fiveprime-species\n",
      "BERT-ADN-combined\n",
      "BERT-ADN-combined-species\n",
      "3-mer-fiveprime\n",
      "4-mer-fiveprime\n",
      "5-mer-fiveprime\n",
      "3-mer-threeprime\n",
      "4-mer-threeprime\n",
      "5-mer-threeprime\n",
      "3-mer-combined\n",
      "4-mer-combined\n",
      "5-mer-combined\n"
     ]
    }
   ],
   "source": [
    "rows = []\n",
    "df_list = []\n",
    "test_predictions = []\n",
    "np.random.seed(42)\n",
    "import copy \n",
    "\n",
    "pipe_index_dict = dict()\n",
    "\n",
    "for organism in set(microarray[\"Organism\"]):\n",
    "    print (organism)\n",
    "    for condition in set(microarray[\"Condition\"]):\n",
    "        print(condition)\n",
    "        micro_df = microarray.query('Condition == @condition and Organism == @organism').reset_index(drop=True)\n",
    "        if len(micro_df) == 0:\n",
    "            continue\n",
    "        y_obs = np.array(micro_df[\"Expression\"])\n",
    "        # compute features\n",
    "        X_bert_five = np.stack(micro_df[\"five_prime_embedding\"],axis=0)\n",
    "        X_bert_five_species = np.stack(micro_df[\"five_prime_embedding_species\"],axis=0)\n",
    "        X_bert_three = np.stack(micro_df[\"three_prime_embedding\"],axis=0)\n",
    "        X_bert_three_species = np.stack(micro_df[\"three_prime_embedding_species\"],axis=0)\n",
    "        X_bert_combined = np.concatenate([X_bert_five,X_bert_three],axis=1)\n",
    "        X_bert_combined_species = np.concatenate([X_bert_five_species,X_bert_three_species],axis=1)\n",
    "\n",
    "        X_3mer_five = np.stack(micro_df['five_prime_seq'].apply(lambda x: kmerizer3.kmerize(x)))\n",
    "        X_4mer_five = np.stack(micro_df['five_prime_seq'].apply(lambda x: kmerizer4.kmerize(x)))\n",
    "        X_5mer_five = np.stack(micro_df['five_prime_seq'].apply(lambda x: kmerizer5.kmerize(x)))\n",
    "        #X_6mer_five = np.stack(micro_df['five_prime_seq'].apply(lambda x: kmerizer6.kmerize(x)))\n",
    "        X_3mer_three = np.stack(micro_df['three_prime_seq'].apply(lambda x: kmerizer3.kmerize(x)))\n",
    "        X_4mer_three = np.stack(micro_df['three_prime_seq'].apply(lambda x: kmerizer4.kmerize(x)))\n",
    "        X_5mer_three = np.stack(micro_df['three_prime_seq'].apply(lambda x: kmerizer5.kmerize(x)))\n",
    "        #X_6mer_three = np.stack(micro_df['three_prime_seq'].apply(lambda x: kmerizer6.kmerize(x)))\n",
    "        X_3mer_combined = np.concatenate([X_3mer_five, X_3mer_three],axis=1)\n",
    "        X_4mer_combined = np.concatenate([X_4mer_five, X_4mer_three],axis=1)\n",
    "        X_5mer_combined = np.concatenate([X_5mer_five, X_5mer_three],axis=1)\n",
    "        #X_6mer_combined = np.concatenate([X_6mer_five, X_6mer_three],axis=1)\n",
    "\n",
    "        data_matrices = {\"BERT-ADN-threeprime\":X_bert_three,\n",
    "                        \"BERT-ADN-threeprime-species\":X_bert_three_species,\n",
    "                        \"BERT-ADN-fiveprime\":X_bert_five,\n",
    "                        \"BERT-ADN-fiveprime-species\":X_bert_five_species,\n",
    "                        \"BERT-ADN-combined\":X_bert_combined,\n",
    "                        \"BERT-ADN-combined-species\":X_bert_combined_species,\n",
    "                         \"3-mer-fiveprime\":X_3mer_five, \n",
    "                         \"4-mer-fiveprime\":X_4mer_five,\n",
    "                         \"5-mer-fiveprime\":X_5mer_five, \n",
    "                         \"3-mer-threeprime\":X_3mer_three,\n",
    "                         \"4-mer-threeprime\":X_4mer_three,\n",
    "                         \"5-mer-threeprime\":X_5mer_three,\n",
    "                         \"3-mer-combined\":X_3mer_combined,\n",
    "                         \"4-mer-combined\":X_4mer_combined,\n",
    "                         \"5-mer-combined\":X_5mer_combined,\n",
    "        }\n",
    "\n",
    "        kfold = sklearn.model_selection.KFold(n_splits=10, shuffle=True, random_state=42)\n",
    "\n",
    "        for key in data_matrices:\n",
    "            print(key)\n",
    "            y = y_obs\n",
    "            X = data_matrices[key]\n",
    "\n",
    "            y_pred_list = []\n",
    "            y_true_list = []\n",
    "            for i, (train_index, test_index) in enumerate(kfold.split(X, y)):\n",
    "                pipe = pipeline.make_pipeline(sklearn.preprocessing.StandardScaler(), sklearn.linear_model.RidgeCV(cv=3, alphas=[10, 100, 1000]))\n",
    "                X_train = X[train_index]\n",
    "                y_train = y[train_index]\n",
    "                X_test = X[test_index]\n",
    "                y_test = y[test_index]\n",
    "                pipe.fit(X_train,y_train)\n",
    "                y_pred = pipe.predict(X_test)\n",
    "                r2_test = pearson_r2_metric(y_test, y_pred)\n",
    "                y_pred_list.append(y_pred)\n",
    "                y_true_list.append(y_test)\n",
    "                rows.append({\"model\":key,\n",
    "                             \"species\":organism,\n",
    "                             \"condition\":condition,\n",
    "                             \"r2\":r2_test})\n",
    "            test_predictions.append(pd.DataFrame({\"model\":key,\n",
    "                                                  \"species\":organism,\n",
    "                                                  \"condition\":condition,\n",
    "                                     \"pred\": np.concatenate(y_pred_list), \n",
    "                                     \"true\": np.concatenate(y_true_list)}))\n",
    "metrics_micro_other = pd.DataFrame(rows)\n",
    "pred_micro_other = pd.concat(test_predictions)\n",
    "pred_micro_other[\"residual\"] = (pred_micro_other[\"true\"] - pred_micro_other[\"pred\"])**2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "620d6e2a-0d71-4a3b-86cd-4863274b7f11",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "WilcoxonResult(statistic=150382582.0, pvalue=2.3461313127918796e-26)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_micro_other\n",
    "scipy.stats.wilcoxon(pred_micro_other.query(\"species == 'Candida glabrata'\").query('model == \"BERT-ADN-combined-species\"')[\"residual\"], pred_micro_other.query(\"species == 'Candida glabrata'\").query('model == \"BERT-ADN-combined\"')[\"residual\"])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "cc933c0b-0588-4549-864e-fbca5d7154c0",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Candida glabrata WilcoxonResult(statistic=150382582.0, pvalue=2.3461313127918796e-26)\n"
     ]
    }
   ],
   "source": [
    "for species in pred_micro_other.species.unique():\n",
    "    sub_micro = pred_micro_other.loc[pred_micro_other.species == species]\n",
    "    print (species, scipy.stats.wilcoxon(sub_micro.query('model == \"BERT-ADN-combined-species\"')[\"residual\"], sub_micro.query('model == \"BERT-ADN-combined\"')[\"residual\"]))\n",
    "    #print (species, scipy.stats.wilcoxon(sub_micro.query('model == \"BERT-ADN-threeprime-species\"')[\"residual\"], sub_micro.query('model == \"BERT-ADN-threeprime\"')[\"residual\"]))\n",
    "    #print (species, scipy.stats.wilcoxon(sub_micro.query('model == \"BERT-ADN-fiveprime-species\"')[\"residual\"], sub_micro.query('model == \"BERT-ADN-fiveprime\"')[\"residual\"]))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "bf0a1215-a42f-4d50-b636-51097bcd6a39",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>r2</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>species</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.456447</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        r2\n",
       "species                   \n",
       "Candida glabrata  0.456447"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics_micro_other.query(\"model == 'BERT-ADN-combined-species'\").groupby([\"species\"])[[\"r2\"]].median().tail(60).sort_values('r2')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "02646053-00fd-4369-a24d-8b77de1446b0",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>r2</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>model</th>\n",
       "      <th>species</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5-mer-threeprime</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.021593</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3-mer-threeprime</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.024481</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4-mer-threeprime</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.025314</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BERT-ADN-threeprime</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.063825</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BERT-ADN-threeprime-species</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.066696</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3-mer-fiveprime</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.107703</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3-mer-combined</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.123278</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4-mer-fiveprime</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.154971</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4-mer-combined</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.162162</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5-mer-combined</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.181400</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5-mer-fiveprime</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.200339</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BERT-ADN-combined</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.425320</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BERT-ADN-fiveprime</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.439048</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BERT-ADN-combined-species</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.456447</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BERT-ADN-fiveprime-species</th>\n",
       "      <th>Candida glabrata</th>\n",
       "      <td>0.462985</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                    r2\n",
       "model                       species                   \n",
       "5-mer-threeprime            Candida glabrata  0.021593\n",
       "3-mer-threeprime            Candida glabrata  0.024481\n",
       "4-mer-threeprime            Candida glabrata  0.025314\n",
       "BERT-ADN-threeprime         Candida glabrata  0.063825\n",
       "BERT-ADN-threeprime-species Candida glabrata  0.066696\n",
       "3-mer-fiveprime             Candida glabrata  0.107703\n",
       "3-mer-combined              Candida glabrata  0.123278\n",
       "4-mer-fiveprime             Candida glabrata  0.154971\n",
       "4-mer-combined              Candida glabrata  0.162162\n",
       "5-mer-combined              Candida glabrata  0.181400\n",
       "5-mer-fiveprime             Candida glabrata  0.200339\n",
       "BERT-ADN-combined           Candida glabrata  0.425320\n",
       "BERT-ADN-fiveprime          Candida glabrata  0.439048\n",
       "BERT-ADN-combined-species   Candida glabrata  0.456447\n",
       "BERT-ADN-fiveprime-species  Candida glabrata  0.462985"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics_micro_other.groupby(['model','species',])[[\"r2\"]].median().query(\"species == 'Candida glabrata'\").sort_values(\"r2\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f197778c-7dc4-4da1-8a77-6acfc415e600",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:anaconda-motifnew]",
   "language": "python",
   "name": "conda-env-anaconda-motifnew-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}