Tutorials/Model_Training_and_Prediction.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "788957bb",
   "metadata": {},
   "source": [
    "## Pair Generation"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0bd7bd33",
   "metadata": {},
   "source": [
    "Generating pairs.    \n",
    "Prefs[0] = x1 descriptors     \n",
    "Prefs[1] = x2 descriptors    \n",
    "Prefs[2] = y1  \n",
    "Prefs[3] = y2 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "fa9d1e38",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import pickle\n",
    "\n",
    "from DPDI.utilities.pairs_generator import PairsGenerator\n",
    "from DPDI.IntegNet import IntegNet\n",
    "\n",
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "de365e21",
   "metadata": {},
   "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>smiles</th>\n",
       "      <th>value</th>\n",
       "      <th>assay</th>\n",
       "      <th>MaxEStateIndex</th>\n",
       "      <th>MinEStateIndex</th>\n",
       "      <th>MinAbsEStateIndex</th>\n",
       "      <th>qed</th>\n",
       "      <th>MolWt</th>\n",
       "      <th>MaxPartialCharge</th>\n",
       "      <th>MinPartialCharge</th>\n",
       "      <th>...</th>\n",
       "      <th>fr_piperzine</th>\n",
       "      <th>fr_priamide</th>\n",
       "      <th>fr_pyridine</th>\n",
       "      <th>fr_sulfide</th>\n",
       "      <th>fr_sulfone</th>\n",
       "      <th>fr_term_acetylene</th>\n",
       "      <th>fr_thiazole</th>\n",
       "      <th>fr_thiophene</th>\n",
       "      <th>fr_unbrch_alkane</th>\n",
       "      <th>fr_urea</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>N=C(N)c1cccc(CN2CC[C@H](NS(=O)(=O)c3ccc(-c4ccn...</td>\n",
       "      <td>0.477024</td>\n",
       "      <td>CHEMBL3885768</td>\n",
       "      <td>0.240192</td>\n",
       "      <td>0.257329</td>\n",
       "      <td>0.115358</td>\n",
       "      <td>0.343176</td>\n",
       "      <td>0.231894</td>\n",
       "      <td>0.028575</td>\n",
       "      <td>0.785021</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>N=C(N)c1cccc(CN2CC[C@H](NS(=O)(=O)c3ccc(-c4ccn...</td>\n",
       "      <td>0.490153</td>\n",
       "      <td>CHEMBL3885772</td>\n",
       "      <td>0.240192</td>\n",
       "      <td>0.257329</td>\n",
       "      <td>0.115358</td>\n",
       "      <td>0.343176</td>\n",
       "      <td>0.231894</td>\n",
       "      <td>0.028575</td>\n",
       "      <td>0.785021</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>N=C(N)c1cccc(CN2CC[C@H](NS(=O)(=O)c3ccc(-c4ccn...</td>\n",
       "      <td>0.446389</td>\n",
       "      <td>CHEMBL3885774</td>\n",
       "      <td>0.240192</td>\n",
       "      <td>0.257329</td>\n",
       "      <td>0.115358</td>\n",
       "      <td>0.343176</td>\n",
       "      <td>0.231894</td>\n",
       "      <td>0.028575</td>\n",
       "      <td>0.785021</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>N=C(N)c1cccc(CN2CC[C@H](NS(=O)(=O)c3ccc(-c4ccn...</td>\n",
       "      <td>0.452954</td>\n",
       "      <td>CHEMBL3885775</td>\n",
       "      <td>0.240192</td>\n",
       "      <td>0.257329</td>\n",
       "      <td>0.115358</td>\n",
       "      <td>0.343176</td>\n",
       "      <td>0.231894</td>\n",
       "      <td>0.028575</td>\n",
       "      <td>0.785021</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>O=C1[C@@H](NS(=O)(=O)c2cc3ncccc3s2)CCN1Cc1cc2c...</td>\n",
       "      <td>0.527352</td>\n",
       "      <td>CHEMBL3885768</td>\n",
       "      <td>0.220448</td>\n",
       "      <td>0.262623</td>\n",
       "      <td>0.488022</td>\n",
       "      <td>0.524445</td>\n",
       "      <td>0.145951</td>\n",
       "      <td>0.028822</td>\n",
       "      <td>0.878645</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 132 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                              smiles     value          assay  \\\n",
       "0  N=C(N)c1cccc(CN2CC[C@H](NS(=O)(=O)c3ccc(-c4ccn...  0.477024  CHEMBL3885768   \n",
       "1  N=C(N)c1cccc(CN2CC[C@H](NS(=O)(=O)c3ccc(-c4ccn...  0.490153  CHEMBL3885772   \n",
       "2  N=C(N)c1cccc(CN2CC[C@H](NS(=O)(=O)c3ccc(-c4ccn...  0.446389  CHEMBL3885774   \n",
       "3  N=C(N)c1cccc(CN2CC[C@H](NS(=O)(=O)c3ccc(-c4ccn...  0.452954  CHEMBL3885775   \n",
       "4  O=C1[C@@H](NS(=O)(=O)c2cc3ncccc3s2)CCN1Cc1cc2c...  0.527352  CHEMBL3885768   \n",
       "\n",
       "   MaxEStateIndex  MinEStateIndex  MinAbsEStateIndex       qed     MolWt  \\\n",
       "0        0.240192        0.257329           0.115358  0.343176  0.231894   \n",
       "1        0.240192        0.257329           0.115358  0.343176  0.231894   \n",
       "2        0.240192        0.257329           0.115358  0.343176  0.231894   \n",
       "3        0.240192        0.257329           0.115358  0.343176  0.231894   \n",
       "4        0.220448        0.262623           0.488022  0.524445  0.145951   \n",
       "\n",
       "   MaxPartialCharge  MinPartialCharge  ...  fr_piperzine  fr_priamide  \\\n",
       "0          0.028575          0.785021  ...           0.0          0.0   \n",
       "1          0.028575          0.785021  ...           0.0          0.0   \n",
       "2          0.028575          0.785021  ...           0.0          0.0   \n",
       "3          0.028575          0.785021  ...           0.0          0.0   \n",
       "4          0.028822          0.878645  ...           0.0          0.0   \n",
       "\n",
       "   fr_pyridine  fr_sulfide  fr_sulfone  fr_term_acetylene  fr_thiazole  \\\n",
       "0          0.5         0.0         0.0                0.0          0.0   \n",
       "1          0.5         0.0         0.0                0.0          0.0   \n",
       "2          0.5         0.0         0.0                0.0          0.0   \n",
       "3          0.5         0.0         0.0                0.0          0.0   \n",
       "4          1.0         0.0         0.0                0.0          0.0   \n",
       "\n",
       "   fr_thiophene  fr_unbrch_alkane  fr_urea  \n",
       "0           1.0               0.0      0.0  \n",
       "1           1.0               0.0      0.0  \n",
       "2           1.0               0.0      0.0  \n",
       "3           1.0               0.0      0.0  \n",
       "4           1.0               0.0      0.0  \n",
       "\n",
       "[5 rows x 132 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv('Data/chembl_fax_rdkit1d_normalized.csv')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "f8ec5db8",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CHEMBL3885768 55\n",
      "CHEMBL3885772 55\n",
      "CHEMBL3885774 55\n",
      "CHEMBL3885775 56\n",
      "CHEMBL968695 55\n",
      "CHEMBL659609 62\n",
      "CHEMBL885070 46\n",
      "CHEMBL3887092 85\n"
     ]
    }
   ],
   "source": [
    "#Check assay ID and the number of molecules in each group.\n",
    "for assay in df['assay'].unique():\n",
    "    print (assay,len(df[df['assay']==assay]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "332218a3",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The number of identical assay group is: 8\n",
      "The length of total prefs is: 13821\n"
     ]
    }
   ],
   "source": [
    "#Involve all molecules as training group for pairs generation. Assay with only 1 molecule will be dropped.\n",
    "PG = PairsGenerator(df)\n",
    "prefs = PG.generate()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "00910b19",
   "metadata": {},
   "source": [
    "## Model Training"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d0389616",
   "metadata": {},
   "source": [
    "Train set: training the model.    \n",
    "Val set: tune hyperparameters.     \n",
    "Test set: verify performances.   \n",
    "The ndcg ranking score requires a true ranking of the input list. To calculate the ndcg, we need keep the predicted molecules in same assay group.     \n",
    "In actual use, input all data as training set for predicting unknown molecules after tuning hyperparameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "49fe76c2",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_set,val_set = train_test_split(df[df['assay'] != 'CHEMBL3887092'])\n",
    "test_set = df[df['assay'] == 'CHEMBL3887092']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "406ebe62",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The number of identical assay group is: 7\n",
      "The length of total prefs is: 5723\n",
      "The number of identical assay group is: 7\n",
      "The length of total prefs is: 671\n"
     ]
    }
   ],
   "source": [
    "PG = PairsGenerator(train_set)\n",
    "train_prefs = PG.generate()\n",
    "PG = PairsGenerator(val_set)\n",
    "val_prefs = PG.generate()\n",
    "#PG.x,PG.y are descriptors and pchembl value of val_set dataframe in a numpy array form."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "b1a80424",
   "metadata": {},
   "outputs": [],
   "source": [
    "#load settings\n",
    "#example is 2-hidden-layer neural network with SGD optimizer(Adam available)\n",
    "#number of input equals descriptors length, set number of neurons in layer 1 and 2 \n",
    "#cuda default is None for cpu training\n",
    "model = IntegNet(129,512,128,optim='SGD',batch_size= 256,n_epoch=200,learning_rate=0.005,cuda='cuda:0')\n",
    "model.load_train_pair(train_prefs)\n",
    "model.load_val_pair(val_prefs,PG.x,PG.y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "e3e6c241",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|█████████████████████████████████████████████████████████████████████████████████| 200/200 [04:45<00:00,  1.43s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Final Train loss is: 0.32201493434284045 val loss is: 0.5498783588409424\n",
      "Final Train NDCG is: 0.9772216968119989 Final val NDCG is: 0.90027395683222\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 1 epoch is defined as training all pairs instead of all molecules.\n",
    "# defining number of epoch and learning rate has higher priority\n",
    "# set n_epochs_stop for early stop\n",
    "# add n_epochs for better model\n",
    "model.train_pair(ep=200,lr=0.01,n_epochs_stop=30,printres=True,plotres=True,printloss=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cbb9ba4f",
   "metadata": {},
   "source": [
    "## Save and load DPDI Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "f9a0abbc",
   "metadata": {},
   "outputs": [],
   "source": [
    "modelname = 'model/finalized_model.pkl'\n",
    "pickle.dump(model, open(modelname, 'wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "ae2fcf38",
   "metadata": {},
   "outputs": [],
   "source": [
    "loaded_model = pickle.load(open(modelname, 'rb'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "08865c9d",
   "metadata": {},
   "source": [
    "## Predicting new molecules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "8a4dd170",
   "metadata": {},
   "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>smiles</th>\n",
       "      <th>value</th>\n",
       "      <th>assay</th>\n",
       "      <th>MaxEStateIndex</th>\n",
       "      <th>MinEStateIndex</th>\n",
       "      <th>MinAbsEStateIndex</th>\n",
       "      <th>qed</th>\n",
       "      <th>MolWt</th>\n",
       "      <th>MaxPartialCharge</th>\n",
       "      <th>MinPartialCharge</th>\n",
       "      <th>...</th>\n",
       "      <th>fr_piperzine</th>\n",
       "      <th>fr_priamide</th>\n",
       "      <th>fr_pyridine</th>\n",
       "      <th>fr_sulfide</th>\n",
       "      <th>fr_sulfone</th>\n",
       "      <th>fr_term_acetylene</th>\n",
       "      <th>fr_thiazole</th>\n",
       "      <th>fr_thiophene</th>\n",
       "      <th>fr_unbrch_alkane</th>\n",
       "      <th>fr_urea</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>153</th>\n",
       "      <td>CCOc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS...</td>\n",
       "      <td>0.343545</td>\n",
       "      <td>CHEMBL3887092</td>\n",
       "      <td>0.454429</td>\n",
       "      <td>0.177146</td>\n",
       "      <td>0.023760</td>\n",
       "      <td>0.260544</td>\n",
       "      <td>0.707041</td>\n",
       "      <td>0.257546</td>\n",
       "      <td>0.417948</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154</th>\n",
       "      <td>COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC(F)(F)F)N1CC...</td>\n",
       "      <td>0.013129</td>\n",
       "      <td>CHEMBL3887092</td>\n",
       "      <td>0.459838</td>\n",
       "      <td>0.005852</td>\n",
       "      <td>0.012673</td>\n",
       "      <td>0.192049</td>\n",
       "      <td>0.872378</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.505134</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>155</th>\n",
       "      <td>COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1F)N1CC[C@H](NS...</td>\n",
       "      <td>0.059081</td>\n",
       "      <td>CHEMBL3887092</td>\n",
       "      <td>0.852523</td>\n",
       "      <td>0.188579</td>\n",
       "      <td>0.151821</td>\n",
       "      <td>0.368090</td>\n",
       "      <td>0.567230</td>\n",
       "      <td>0.264054</td>\n",
       "      <td>0.505134</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>156</th>\n",
       "      <td>COc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS(...</td>\n",
       "      <td>0.776805</td>\n",
       "      <td>CHEMBL3887092</td>\n",
       "      <td>0.333091</td>\n",
       "      <td>0.211611</td>\n",
       "      <td>0.127987</td>\n",
       "      <td>0.330934</td>\n",
       "      <td>0.455608</td>\n",
       "      <td>0.257546</td>\n",
       "      <td>0.408333</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>157</th>\n",
       "      <td>CCOC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC)N1CC[C@H](...</td>\n",
       "      <td>0.407002</td>\n",
       "      <td>CHEMBL3887092</td>\n",
       "      <td>0.408238</td>\n",
       "      <td>0.215740</td>\n",
       "      <td>0.173094</td>\n",
       "      <td>0.313816</td>\n",
       "      <td>0.541552</td>\n",
       "      <td>0.264855</td>\n",
       "      <td>0.408333</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 132 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                smiles     value  \\\n",
       "153  CCOc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS...  0.343545   \n",
       "154  COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC(F)(F)F)N1CC...  0.013129   \n",
       "155  COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1F)N1CC[C@H](NS...  0.059081   \n",
       "156  COc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS(...  0.776805   \n",
       "157  CCOC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC)N1CC[C@H](...  0.407002   \n",
       "\n",
       "             assay  MaxEStateIndex  MinEStateIndex  MinAbsEStateIndex  \\\n",
       "153  CHEMBL3887092        0.454429        0.177146           0.023760   \n",
       "154  CHEMBL3887092        0.459838        0.005852           0.012673   \n",
       "155  CHEMBL3887092        0.852523        0.188579           0.151821   \n",
       "156  CHEMBL3887092        0.333091        0.211611           0.127987   \n",
       "157  CHEMBL3887092        0.408238        0.215740           0.173094   \n",
       "\n",
       "          qed     MolWt  MaxPartialCharge  MinPartialCharge  ...  \\\n",
       "153  0.260544  0.707041          0.257546          0.417948  ...   \n",
       "154  0.192049  0.872378          1.000000          0.505134  ...   \n",
       "155  0.368090  0.567230          0.264054          0.505134  ...   \n",
       "156  0.330934  0.455608          0.257546          0.408333  ...   \n",
       "157  0.313816  0.541552          0.264855          0.408333  ...   \n",
       "\n",
       "     fr_piperzine  fr_priamide  fr_pyridine  fr_sulfide  fr_sulfone  \\\n",
       "153           0.0          0.0          0.5         0.0         0.0   \n",
       "154           0.0          0.0          0.5         0.0         0.0   \n",
       "155           0.0          0.0          1.0         0.0         0.0   \n",
       "156           0.0          0.0          1.0         0.0         0.0   \n",
       "157           0.0          0.0          1.0         0.0         0.0   \n",
       "\n",
       "     fr_term_acetylene  fr_thiazole  fr_thiophene  fr_unbrch_alkane  fr_urea  \n",
       "153                0.0          1.0           0.0               0.0      0.0  \n",
       "154                0.0          1.0           0.0               0.0      0.0  \n",
       "155                0.0          0.0           0.0               0.0      0.0  \n",
       "156                0.0          0.0           0.0               0.0      0.0  \n",
       "157                0.0          0.0           0.0               0.0      0.0  \n",
       "\n",
       "[5 rows x 132 columns]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# if the test set has not been normalized before, use before saved Minmax value of each column to scale it. \n",
    "# from DPDI.utilities.feature_engineering import reverse\n",
    "# test_set = reverse(test_set)\n",
    "test_set.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "477e02c8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[-1.4002954   0.09351856 -0.3855068  -2.9241128  -1.0871704   0.91476023\n",
      " -1.1158445  -2.0251615   1.4911327  -1.8385124  -2.9240832  -1.4707011\n",
      " -1.4001238  -1.8525149  -0.12492952 -1.304274   -2.2296748  -0.7600633\n",
      " -0.65520084 -0.383196   -2.7950428  -1.6985021  -1.7217685   0.35735464\n",
      " -1.190069   -1.3090426  -1.9579406  -1.4211136  -1.9111372  -1.8631464\n",
      " -2.3930776  -1.1035527  -1.2442901   0.21513852 -0.32485282 -0.7851641\n",
      " -1.0291455  -2.3035717  -1.11591    -0.83119416 -2.076385   -1.5394866\n",
      " -1.9641085  -1.3872988   0.16114798 -0.4654739  -0.19083081 -2.3568783\n",
      " -2.0704052  -0.9819641  -1.9025068  -2.015341   -2.6128254   0.43415385\n",
      " -0.7378543  -1.7306882  -1.8069602  -0.37471712 -1.8092103  -2.2462115\n",
      " -2.271363    0.3936357  -2.3821874  -2.2479188  -0.26088667 -2.4680681\n",
      " -0.3208356  -2.6856937  -1.0633312  -0.57688296 -0.31276    -0.8492989\n",
      " -2.4294758  -0.8957543  -1.0501616  -1.7523451  -0.07823297 -1.832142\n",
      " -2.971512   -0.70899326 -2.3290794  -0.06025294 -0.915375   -2.0267868\n",
      " -2.0036716 ]\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>smiles</th>\n",
       "      <th>pred_value</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>153</th>\n",
       "      <td>CCOc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS...</td>\n",
       "      <td>-1.400295</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154</th>\n",
       "      <td>COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC(F)(F)F)N1CC...</td>\n",
       "      <td>0.093519</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>155</th>\n",
       "      <td>COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1F)N1CC[C@H](NS...</td>\n",
       "      <td>-0.385507</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>156</th>\n",
       "      <td>COc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS(...</td>\n",
       "      <td>-2.924113</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>157</th>\n",
       "      <td>CCOC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC)N1CC[C@H](...</td>\n",
       "      <td>-1.087170</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>310</th>\n",
       "      <td>CN(C)c1ncc(S(=O)(=O)N[C@H]2CCN([C@H](Cc3cc4ccn...</td>\n",
       "      <td>-2.329079</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>311</th>\n",
       "      <td>CCOC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC)N1CC[C@H](...</td>\n",
       "      <td>-0.060253</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>312</th>\n",
       "      <td>COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC(F)(F)F)N1CC...</td>\n",
       "      <td>-0.915375</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>313</th>\n",
       "      <td>CN(C)c1ccc(S(=O)(=O)N[C@H]2CCN([C@H](Cc3cc4ccn...</td>\n",
       "      <td>-2.026787</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>318</th>\n",
       "      <td>COc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS(...</td>\n",
       "      <td>-2.003672</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>85 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                smiles  pred_value\n",
       "153  CCOc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS...   -1.400295\n",
       "154  COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC(F)(F)F)N1CC...    0.093519\n",
       "155  COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1F)N1CC[C@H](NS...   -0.385507\n",
       "156  COc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS(...   -2.924113\n",
       "157  CCOC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC)N1CC[C@H](...   -1.087170\n",
       "..                                                 ...         ...\n",
       "310  CN(C)c1ncc(S(=O)(=O)N[C@H]2CCN([C@H](Cc3cc4ccn...   -2.329079\n",
       "311  CCOC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC)N1CC[C@H](...   -0.060253\n",
       "312  COC(=O)[C@@H](Cc1cc2ccnc(N)c2cc1OC(F)(F)F)N1CC...   -0.915375\n",
       "313  CN(C)c1ccc(S(=O)(=O)N[C@H]2CCN([C@H](Cc3cc4ccn...   -2.026787\n",
       "318  COc1cc2c(N)nccc2cc1C[C@H](C(=O)O)N1CC[C@H](NS(...   -2.003672\n",
       "\n",
       "[85 rows x 2 columns]"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#The prediction is surrogate value only for ranking. Negative number is acceptable. \n",
    "x = np.array(test_set.drop(['smiles','value','assay'],axis=1))\n",
    "pred_y = loaded_model.predict(x)\n",
    "print(pred_y)\n",
    "res = pd.concat([test_set['smiles'],pd.DataFrame({'pred_value':pred_y}).set_index(test_set.index)],axis=1)\n",
    "res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cba82dff",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}