training_demo.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "e6472bf3",
   "metadata": {},
   "source": [
    "# Training Demo Enformer Celltyping\n",
    "\n",
    "This workbook steps through training Enformer Celltyping using the small subset of data available in the repo."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "084c219f",
   "metadata": {},
   "source": [
    "## Step 1 Create Enformer Celltyping model \n",
    "\n",
    "Enformer Celltyping fine-tunes using the enformer model which is available from tensorflow hub and needs to be downloaded (if you ran `python bin/download_Enformer_Celltyping_dependencies.py` from the README it will have been downloaded already). The enformer weights are lifted and added to a recreated enformer model with the final layers after the attention layers removed to make the model more suitable for fine-tuning, removing its task-specific layers but keeping it's DNA-interaciton information.\n",
    "\n",
    "Let's initiate Enformer Celltyping. Training Enformer Celltyping was done in two steps:\n",
    "\n",
    "1. Pretrain a split model\n",
    "Since the enformer model is being used to embed the DNA information in Enformer Celltyping, the weights for the DNA information are already sensibly initiated. To avoid the DNA information dominating the final predictions in Enformer Celltyping (and the model just ignoring the chromatin accessibility information), we first trained Enformer Celltyping with a split architecture as some 'warm-up' steps. For a given genomic position, the model predicts the (1) average signal at this position and the (2) quantile distribution of the signal for the 104 trianing cell types from the DNA information and the (3) delta i.e. difference between the average signal and the cell type-specific signal for the cell type of interest from the chromatin accessibility information. We show how this two-stage training approach improves performance in our manuscript.\n",
    "\n",
    "This warm-up stage is run for 1,000 steps where each step contains the full batch size of 128 positions. \n",
    "\n",
    "Here, for the demo, we will just train for 20 steps on a small dataset but note this will affect overall performance.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ad6e0f91",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-09-08 09:58:48.073014: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudart.so.11.0\n",
      "2023-09-08 09:58:50.899761: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcuda.so.1\n",
      "2023-09-08 09:58:50.946427: E tensorflow/stream_executor/cuda/cuda_driver.cc:328] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected\n",
      "2023-09-08 09:58:50.946447: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:156] kernel driver does not appear to be running on this host (workstation-neurogenomics): /proc/driver/nvidia/version does not exist\n",
      "2023-09-08 09:58:50.946953: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"EnfCelltyping\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "dnaInput (InputLayer)           [(None, 896, 1536)]  0                                            \n",
      "__________________________________________________________________________________________________\n",
      "dense1_quant (Dense)            (None, 896, 1536)    2360832     dnaInput[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "ChromAccessLclInput (InputLayer [(None, 1562)]       0                                            \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_quant (Dropout)        (None, 896, 1536)    0           dense1_quant[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "dense1_dna (Dense)              (None, 896, 1536)    2360832     dnaInput[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "tf.clip_by_value (TFOpLambda)   (None, 1562)         0           ChromAccessLclInput[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_finalpoint_quant (Ba (None, 896, 1536)    6144        dropout1_quant[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_dna (Dropout)          (None, 896, 1536)    0           dense1_dna[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.add (TFOpLambda)        (None, 1562)         0           tf.clip_by_value[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "ChromAccessGblInput (InputLayer [(None, 14592)]      0                                            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_4 (TFOpLambda) (None, 896, 1536)    0           batchnorm1_finalpoint_quant[0][0]\n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_finalpoint (BatchNor (None, 896, 1536)    6144        dropout1_dna[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "128bp_embedding (Embedding)     (None, 1562, 25)     39050       tf.math.add[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "500bp_embedding (Embedding)     (None, 1562, 40)     15640       tf.math.add[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "5kbp_embedding (Embedding)      (None, 1562, 45)     1800        tf.math.add[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "gbl_embedding (Embedding)       (None, 14592, 25)    364800      ChromAccessGblInput[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_2 (TFOpLambda)  (None, 896, 1536)    0           tf.math.multiply_4[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply (TFOpLambda)   (None, 896, 1536)    0           batchnorm1_finalpoint[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "flatten128 (Flatten)            (None, 39050)        0           128bp_embedding[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "flatten500 (Flatten)            (None, 62480)        0           500bp_embedding[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "flatten5k (Flatten)             (None, 70290)        0           5kbp_embedding[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "flattenGbl (Flatten)            (None, 364800)       0           gbl_embedding[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_5 (TFOpLambda) (None, 896, 1536)    0           tf.math.sigmoid_2[0][0]          \n",
      "                                                                 batchnorm1_finalpoint_quant[0][0]\n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid (TFOpLambda)    (None, 896, 1536)    0           tf.math.multiply[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "cat_embed (Concatenate)         (None, 536620)       0           flatten128[0][0]                 \n",
      "                                                                 flatten500[0][0]                 \n",
      "                                                                 flatten5k[0][0]                  \n",
      "                                                                 flattenGbl[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_finalpoint_quant (Conv1 (None, 896, 3072)    4721664     tf.math.multiply_5[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_1 (TFOpLambda) (None, 896, 1536)    0           tf.math.sigmoid[0][0]            \n",
      "                                                                 batchnorm1_finalpoint[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "dense_combn1_celltype (Dense)   (None, 3072)         1648499712  cat_embed[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_finalpoint_quant (Drop (None, 896, 3072)    0           conv1d1_finalpoint_quant[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_finalpoint (Conv1D)     (None, 896, 3072)    4721664     tf.math.multiply_1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_celltype (Dropout)     (None, 3072)         0           dense_combn1_celltype[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_6 (TFOpLambda) (None, 896, 3072)    0           dropout1_finalpoint_quant[0][0]  \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_finalpoint (Dropout)   (None, 896, 3072)    0           conv1d1_finalpoint[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "dense_combn2_celltype (Dense)   (None, 3072)         9440256     dropout1_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_3 (TFOpLambda)  (None, 896, 3072)    0           tf.math.multiply_6[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_2 (TFOpLambda) (None, 896, 3072)    0           dropout1_finalpoint[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "dropout2_celltype (Dropout)     (None, 3072)         0           dense_combn2_celltype[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_7 (TFOpLambda) (None, 896, 3072)    0           tf.math.sigmoid_3[0][0]          \n",
      "                                                                 dropout1_finalpoint_quant[0][0]  \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_1 (TFOpLambda)  (None, 896, 3072)    0           tf.math.multiply_2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_1 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_2 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_3 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_4 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_5 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_6 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_1 (Dense)        (None, 896, 10)      30730       tf.math.multiply_7[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_2 (Dense)        (None, 896, 10)      30730       tf.math.multiply_7[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_3 (Dense)        (None, 896, 10)      30730       tf.math.multiply_7[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_4 (Dense)        (None, 896, 10)      30730       tf.math.multiply_7[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_5 (Dense)        (None, 896, 10)      30730       tf.math.multiply_7[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_6 (Dense)        (None, 896, 10)      30730       tf.math.multiply_7[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_3 (TFOpLambda) (None, 896, 3072)    0           tf.math.sigmoid_1[0][0]          \n",
      "                                                                 dropout1_finalpoint[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims (TFOpLambda)     (None, 896, 1)       0           dense_chan_celltype_1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_1 (TFOpLambda)   (None, 896, 1)       0           dense_chan_celltype_2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_2 (TFOpLambda)   (None, 896, 1)       0           dense_chan_celltype_3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_3 (TFOpLambda)   (None, 896, 1)       0           dense_chan_celltype_4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_4 (TFOpLambda)   (None, 896, 1)       0           dense_chan_celltype_5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_5 (TFOpLambda)   (None, 896, 1)       0           dense_chan_celltype_6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_6 (TFOpLambda)   (None, 896, 1, 10)   0           quant_out_chan_1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_7 (TFOpLambda)   (None, 896, 1, 10)   0           quant_out_chan_2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_8 (TFOpLambda)   (None, 896, 1, 10)   0           quant_out_chan_3[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_9 (TFOpLambda)   (None, 896, 1, 10)   0           quant_out_chan_4[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_10 (TFOpLambda)  (None, 896, 1, 10)   0           quant_out_chan_5[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_11 (TFOpLambda)  (None, 896, 1, 10)   0           quant_out_chan_6[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "avg (Dense)                     (None, 896, 6)       18438       tf.math.multiply_3[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "delta (Concatenate)             (None, 896, 6)       0           tf.expand_dims[0][0]             \n",
      "                                                                 tf.expand_dims_1[0][0]           \n",
      "                                                                 tf.expand_dims_2[0][0]           \n",
      "                                                                 tf.expand_dims_3[0][0]           \n",
      "                                                                 tf.expand_dims_4[0][0]           \n",
      "                                                                 tf.expand_dims_5[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "quant (Concatenate)             (None, 896, 6, 10)   0           tf.expand_dims_6[0][0]           \n",
      "                                                                 tf.expand_dims_7[0][0]           \n",
      "                                                                 tf.expand_dims_8[0][0]           \n",
      "                                                                 tf.expand_dims_9[0][0]           \n",
      "                                                                 tf.expand_dims_10[0][0]          \n",
      "                                                                 tf.expand_dims_11[0][0]          \n",
      "==================================================================================================\n",
      "Total params: 1,689,261,804\n",
      "Trainable params: 1,689,255,660\n",
      "Non-trainable params: 6,144\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "#create Enformer Celltyping - for training\n",
    "from EnformerCelltyping.enf_celltyping import Enformer_Celltyping\n",
    "from EnformerCelltyping.utils import pearsonR\n",
    "import tensorflow as tf\n",
    "\n",
    "assays = ['h3k27ac', 'h3k4me1', 'h3k4me3', 'h3k9me3', 'h3k27me3', 'h3k36me3']\n",
    "learning_rate = 0.0002\n",
    "\n",
    "#Using all the model default parameters for the architecture\n",
    "#Set use_prebuilt_model to False since we want to train \n",
    "#Enformer Celltyping from scratch\n",
    "model = Enformer_Celltyping(assays=assays,\n",
    "                            use_prebuilt_model=False,\n",
    "                            model_arch = 'split') #model_arch = split creates the architecture for the pre-training\n",
    "#compile the model, the model is separated into 3 channels: \n",
    "# 1. DNA channel which predicts an average histone mark score across\n",
    "#    all trainig cell types\n",
    "# 2. The DNA channel is also used to predict average quantile distribution\n",
    "#    histone mark score across all trainig cell types. We show that adding \n",
    "#    this output improved overall performance in our manuscript.\n",
    "# 3. Chromatin accessibility channel which predicts the delta between\n",
    "#    the average histone mark value for that region and the cell type-\n",
    "#    specific one.\n",
    "# Thus there are 3 loss functions\n",
    "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=learning_rate),\n",
    "          loss={'avg':tf.keras.losses.poisson,\n",
    "                'delta':tf.keras.losses.mean_squared_error,\n",
    "                'quant':tf.keras.losses.categorical_crossentropy,#since a distribution (actually 6 dists)\n",
    "                },\n",
    "          metrics=['mse',pearsonR])\n",
    "\n",
    "#Let's view the model\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b6ed785a",
   "metadata": {},
   "source": [
    "This model does not include the layers of enformer since the DNA for the demo regions\n",
    "has already been passed through the pre-trained enformer model. Since the layers are frozen \n",
    "these weights will not update so we can save on compute time and RAM by pre-running it. If\n",
    "you want to build a version of Enformer Celltyping with the enformer layers run:\n",
    "\n",
    "```\n",
    "from EnformerCelltyping.constants import DATA_PATH\n",
    "from EnformerCelltyping.enf_celltyping import Enformer_Celltyping\n",
    "\n",
    "assays = ['h3k27ac', 'h3k4me1', 'h3k4me3', 'h3k9me3', 'h3k27me3', 'h3k36me3']\n",
    "\n",
    "#Enformer Celltyping from scratch\n",
    "#Include enformer layers\n",
    "model = Enformer_Celltyping(assays=assays,\n",
    "                            use_prebuilt_model=False,\n",
    "                            model_arch = 'split',\n",
    "                            enf_path = str(DATA_PATH / \"enformer_model\"))\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e2a7ad29",
   "metadata": {},
   "source": [
    "## Step 2 Get data loader\n",
    "\n",
    "We have include pre-computed training data with 5 cell types from EpiMap at two \n",
    "locations each in chromosome 10. We will train on these using a tensorflow data\n",
    "loader.\n",
    "\n",
    "If you want more information on the creation of this data, see [full_training_recreation.ipynb](https://github.com/neurogenomics/EnformerCelltyping/blob/master/full_training_recreation.ipynb)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "b9e66bb8",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Now let's get the data ready\n",
    "from EnformerCelltyping.constants import DATA_PATH\n",
    "import glob\n",
    "\n",
    "#get links to precomputed, demo training data\n",
    "#5 cell types from EpiMap at two locations each in chromosome 10\n",
    "train_data = glob.glob(str(DATA_PATH/'demo/*[0-9].npz'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "763bb21e",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Now use the provided custom data loader on this data\n",
    "from EnformerCelltyping.utils import PreSavedDataGen\n",
    "\n",
    "train_dataloader = PreSavedDataGen(files=train_data,\n",
    "                                   batch_size=2,\n",
    "                                   combn_y = False)#since this is the pre-training stage we want 3 output channels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "1362949a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X input to model shape\n",
      "{'dna': TensorShape([2, 896, 1536]), 'chrom_access_lcl': TensorShape([2, 1562]), 'chrom_access_gbl': TensorShape([2, 14592])}\n",
      "Y output from model shape\n",
      "{'avg': TensorShape([2, 896, 6]), 'delta': TensorShape([2, 896, 6]), 'quant': TensorShape([2, 896, 6, 10])}\n"
     ]
    }
   ],
   "source": [
    "#test ouput\n",
    "one_samp = next(iter(train_dataloader))\n",
    "#shape of X inputs\n",
    "print(\"X input to model shape\")\n",
    "print({key: dat.shape for key, dat in one_samp[0].items()})\n",
    "#shape of Y inputs\n",
    "print(\"Y output from model shape\")\n",
    "print({key: dat.shape for key, dat in one_samp[1].items()})\n",
    "#You can see the three output channels for the split pre-training"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a85084ba",
   "metadata": {},
   "source": [
    "## Step 3 Train Enformer Celltyping\n",
    "\n",
    "Now we have our data loader and model ready we can start the pretraining."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "6d3a6a1d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-09-08 09:58:53.054379: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:176] None of the MLIR Optimization Passes are enabled (registered 2)\n",
      "2023-09-08 09:58:53.054801: I tensorflow/core/platform/profile_utils/cpu_utils.cc:114] CPU Frequency: 1996250000 Hz\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 - 7s - loss: 3.4526 - avg_loss: 0.8944 - delta_loss: 0.1434 - quant_loss: 2.4149 - avg_mse: 0.2708 - avg_pearsonR: -3.8901e-02 - delta_mse: 0.1434 - delta_pearsonR: -5.2481e-04 - quant_mse: 0.0376 - quant_pearsonR: -7.1578e-03\n",
      "Epoch 2/20\n",
      "1/1 - 4s - loss: 3.1921 - avg_loss: 0.8686 - delta_loss: 0.2172 - quant_loss: 2.1064 - avg_mse: 0.2414 - avg_pearsonR: 0.2703 - delta_mse: 0.2172 - delta_pearsonR: -6.5134e-03 - quant_mse: 0.0294 - quant_pearsonR: 0.0242\n",
      "Epoch 3/20\n",
      "1/1 - 4s - loss: 2.9514 - avg_loss: 0.8759 - delta_loss: 0.1801 - quant_loss: 1.8954 - avg_mse: 0.1359 - avg_pearsonR: 0.5722 - delta_mse: 0.1801 - delta_pearsonR: 0.1083 - quant_mse: 0.0226 - quant_pearsonR: 0.0242\n",
      "Epoch 4/20\n",
      "1/1 - 4s - loss: 2.7749 - avg_loss: 0.8628 - delta_loss: 0.1683 - quant_loss: 1.7438 - avg_mse: 0.1160 - avg_pearsonR: 0.7106 - delta_mse: 0.1683 - delta_pearsonR: 0.0752 - quant_mse: 0.0200 - quant_pearsonR: -4.8860e-02\n",
      "Epoch 5/20\n",
      "1/1 - 4s - loss: 2.6265 - avg_loss: 0.8304 - delta_loss: 0.2012 - quant_loss: 1.5949 - avg_mse: 0.0708 - avg_pearsonR: 0.6587 - delta_mse: 0.2012 - delta_pearsonR: 0.0165 - quant_mse: 0.0145 - quant_pearsonR: 0.0420\n",
      "Epoch 6/20\n",
      "1/1 - 4s - loss: 2.4535 - avg_loss: 0.8315 - delta_loss: 0.1285 - quant_loss: 1.4934 - avg_mse: 0.1679 - avg_pearsonR: 0.7237 - delta_mse: 0.1285 - delta_pearsonR: 0.1035 - quant_mse: 0.0117 - quant_pearsonR: 0.0293\n",
      "Epoch 7/20\n",
      "1/1 - 4s - loss: 2.3489 - avg_loss: 0.8197 - delta_loss: 0.1571 - quant_loss: 1.3721 - avg_mse: 0.0454 - avg_pearsonR: 0.6963 - delta_mse: 0.1571 - delta_pearsonR: 0.1541 - quant_mse: 0.0087 - quant_pearsonR: 0.0752\n",
      "Epoch 8/20\n",
      "1/1 - 4s - loss: 2.2198 - avg_loss: 0.8213 - delta_loss: 0.1215 - quant_loss: 1.2770 - avg_mse: 0.1370 - avg_pearsonR: 0.8843 - delta_mse: 0.1215 - delta_pearsonR: 0.1545 - quant_mse: 0.0063 - quant_pearsonR: 0.1227\n",
      "Epoch 9/20\n",
      "1/1 - 4s - loss: 2.1829 - avg_loss: 0.8200 - delta_loss: 0.1231 - quant_loss: 1.2398 - avg_mse: 0.0556 - avg_pearsonR: 0.8285 - delta_mse: 0.1231 - delta_pearsonR: 0.2177 - quant_mse: 0.0053 - quant_pearsonR: 0.1506\n",
      "Epoch 10/20\n",
      "1/1 - 4s - loss: 2.2476 - avg_loss: 0.8189 - delta_loss: 0.1819 - quant_loss: 1.2468 - avg_mse: 0.0731 - avg_pearsonR: 0.8803 - delta_mse: 0.1819 - delta_pearsonR: 0.0793 - quant_mse: 0.0056 - quant_pearsonR: 0.1720\n",
      "Epoch 11/20\n",
      "1/1 - 4s - loss: 2.2427 - avg_loss: 0.8231 - delta_loss: 0.1711 - quant_loss: 1.2485 - avg_mse: 0.0499 - avg_pearsonR: 0.7626 - delta_mse: 0.1711 - delta_pearsonR: 0.2919 - quant_mse: 0.0048 - quant_pearsonR: 0.2379\n",
      "Epoch 12/20\n",
      "1/1 - 4s - loss: 2.1052 - avg_loss: 0.8115 - delta_loss: 0.1554 - quant_loss: 1.1383 - avg_mse: 0.0510 - avg_pearsonR: 0.8088 - delta_mse: 0.1554 - delta_pearsonR: 0.1042 - quant_mse: 0.0036 - quant_pearsonR: 0.3215\n",
      "Epoch 13/20\n",
      "1/1 - 4s - loss: 2.2470 - avg_loss: 0.8234 - delta_loss: 0.2539 - quant_loss: 1.1696 - avg_mse: 0.0492 - avg_pearsonR: 0.7174 - delta_mse: 0.2539 - delta_pearsonR: 0.1104 - quant_mse: 0.0054 - quant_pearsonR: 0.0920\n",
      "Epoch 14/20\n",
      "1/1 - 4s - loss: 2.0899 - avg_loss: 0.7981 - delta_loss: 0.1531 - quant_loss: 1.1387 - avg_mse: 0.0305 - avg_pearsonR: 0.9120 - delta_mse: 0.1531 - delta_pearsonR: 0.3303 - quant_mse: 0.0036 - quant_pearsonR: 0.4015\n",
      "Epoch 15/20\n",
      "1/1 - 4s - loss: 2.1633 - avg_loss: 0.8170 - delta_loss: 0.2287 - quant_loss: 1.1176 - avg_mse: 0.0250 - avg_pearsonR: 0.8275 - delta_mse: 0.2287 - delta_pearsonR: 0.1871 - quant_mse: 0.0034 - quant_pearsonR: 0.2753\n",
      "Epoch 16/20\n",
      "1/1 - 4s - loss: 2.1435 - avg_loss: 0.8127 - delta_loss: 0.2301 - quant_loss: 1.1008 - avg_mse: 0.0305 - avg_pearsonR: 0.8640 - delta_mse: 0.2301 - delta_pearsonR: 0.3617 - quant_mse: 0.0030 - quant_pearsonR: 0.3194\n",
      "Epoch 17/20\n",
      "1/1 - 4s - loss: 2.0691 - avg_loss: 0.8188 - delta_loss: 0.1123 - quant_loss: 1.1379 - avg_mse: 0.0497 - avg_pearsonR: 0.8049 - delta_mse: 0.1123 - delta_pearsonR: 0.2233 - quant_mse: 0.0039 - quant_pearsonR: 0.4139\n",
      "Epoch 18/20\n",
      "1/1 - 4s - loss: 1.9708 - avg_loss: 0.8059 - delta_loss: 0.1156 - quant_loss: 1.0493 - avg_mse: 0.0463 - avg_pearsonR: 0.7831 - delta_mse: 0.1156 - delta_pearsonR: 0.4229 - quant_mse: 0.0029 - quant_pearsonR: 0.4688\n",
      "Epoch 19/20\n",
      "1/1 - 4s - loss: 1.9610 - avg_loss: 0.8019 - delta_loss: 0.0933 - quant_loss: 1.0658 - avg_mse: 0.0298 - avg_pearsonR: 0.8004 - delta_mse: 0.0933 - delta_pearsonR: 0.4269 - quant_mse: 0.0030 - quant_pearsonR: 0.5467\n",
      "Epoch 20/20\n",
      "1/1 - 4s - loss: 1.9230 - avg_loss: 0.7944 - delta_loss: 0.1040 - quant_loss: 1.0246 - avg_mse: 0.0179 - avg_pearsonR: 0.8838 - delta_mse: 0.1040 - delta_pearsonR: 0.5160 - quant_mse: 0.0025 - quant_pearsonR: 0.5736\n",
      "Time taken:  0:01:24.185312\n"
     ]
    }
   ],
   "source": [
    "n_steps = 20\n",
    "\n",
    "import datetime\n",
    "strt = datetime.datetime.now()\n",
    "\n",
    "# Train the model\n",
    "model.fit(\n",
    "        train_dataloader,\n",
    "        epochs=n_steps,\n",
    "        verbose=2\n",
    "    )\n",
    "\n",
    "\n",
    "end = datetime.datetime.now()\n",
    "print('Time taken: ',end-strt)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5fe9653a",
   "metadata": {},
   "source": [
    "Now we have sensibly initiated the weights for both the DNA and chromatin accessibility streams, we can combine them to give a cell type-specific prediciton as an output.\n",
    "\n",
    "First let's save these weights to load with the combined architecture:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "dc4c640b",
   "metadata": {},
   "outputs": [],
   "source": [
    "#just use project directory for now, will delete it after\n",
    "from EnformerCelltyping.constants import PROJECT_PATH\n",
    "model.save_weights(PROJECT_PATH / 'test_split_weights.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e9e113ed",
   "metadata": {},
   "source": [
    "Now we can more on to stage 2 of trianing\n",
    "\n",
    "2. Train a combined model\n",
    "Now we have sensibly initiated the weights for both the DNA and chromatin accessibility streams, we can combine them to give a cell type-specific prediciton as an output. For the manuscript, we trained the model for 4 full epochs (6940 steps of 128 batch size - 888,320 positions) stopping when it started to overfit, taking 1.5 days on a 80GB RAM GPU.\n",
    "\n",
    "Here we will just train for 20 steps on the same , small sample data.\n",
    "\n",
    "First, let's create Enformer CElltypign with the combined architecture"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "36b6a826",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"EnformerCelltyping\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "dnaInput (InputLayer)           [(None, 896, 1536)]  0                                            \n",
      "__________________________________________________________________________________________________\n",
      "dense1_quant (Dense)            (None, 896, 1536)    2360832     dnaInput[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "ChromAccessLclInput (InputLayer [(None, 1562)]       0                                            \n",
      "__________________________________________________________________________________________________\n",
      "dense1_dna (Dense)              (None, 896, 1536)    2360832     dnaInput[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_quant (Dropout)        (None, 896, 1536)    0           dense1_quant[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "tf.clip_by_value_1 (TFOpLambda) (None, 1562)         0           ChromAccessLclInput[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_dna (Dropout)          (None, 896, 1536)    0           dense1_dna[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_finalpoint_quant (Ba (None, 896, 1536)    6144        dropout1_quant[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.add_1 (TFOpLambda)      (None, 1562)         0           tf.clip_by_value_1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "ChromAccessGblInput (InputLayer [(None, 14592)]      0                                            \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_finalpoint (BatchNor (None, 896, 1536)    6144        dropout1_dna[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_12 (TFOpLambda (None, 896, 1536)    0           batchnorm1_finalpoint_quant[0][0]\n",
      "__________________________________________________________________________________________________\n",
      "128bp_embedding (Embedding)     (None, 1562, 25)     39050       tf.math.add_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "500bp_embedding (Embedding)     (None, 1562, 40)     15640       tf.math.add_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "5kbp_embedding (Embedding)      (None, 1562, 45)     1800        tf.math.add_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "gbl_embedding (Embedding)       (None, 14592, 25)    364800      ChromAccessGblInput[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_8 (TFOpLambda) (None, 896, 1536)    0           batchnorm1_finalpoint[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_6 (TFOpLambda)  (None, 896, 1536)    0           tf.math.multiply_12[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "flatten128 (Flatten)            (None, 39050)        0           128bp_embedding[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "flatten500 (Flatten)            (None, 62480)        0           500bp_embedding[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "flatten5k (Flatten)             (None, 70290)        0           5kbp_embedding[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "flattenGbl (Flatten)            (None, 364800)       0           gbl_embedding[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_4 (TFOpLambda)  (None, 896, 1536)    0           tf.math.multiply_8[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_13 (TFOpLambda (None, 896, 1536)    0           tf.math.sigmoid_6[0][0]          \n",
      "                                                                 batchnorm1_finalpoint_quant[0][0]\n",
      "__________________________________________________________________________________________________\n",
      "cat_embed (Concatenate)         (None, 536620)       0           flatten128[0][0]                 \n",
      "                                                                 flatten500[0][0]                 \n",
      "                                                                 flatten5k[0][0]                  \n",
      "                                                                 flattenGbl[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_9 (TFOpLambda) (None, 896, 1536)    0           tf.math.sigmoid_4[0][0]          \n",
      "                                                                 batchnorm1_finalpoint[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_finalpoint_quant (Conv1 (None, 896, 3072)    4721664     tf.math.multiply_13[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "dense_combn1_celltype (Dense)   (None, 3072)         1648499712  cat_embed[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_finalpoint (Conv1D)     (None, 896, 3072)    4721664     tf.math.multiply_9[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_finalpoint_quant (Drop (None, 896, 3072)    0           conv1d1_finalpoint_quant[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_celltype (Dropout)     (None, 3072)         0           dense_combn1_celltype[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_finalpoint (Dropout)   (None, 896, 3072)    0           conv1d1_finalpoint[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_14 (TFOpLambda (None, 896, 3072)    0           dropout1_finalpoint_quant[0][0]  \n",
      "__________________________________________________________________________________________________\n",
      "dense_combn2_celltype (Dense)   (None, 3072)         9440256     dropout1_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_10 (TFOpLambda (None, 896, 3072)    0           dropout1_finalpoint[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_7 (TFOpLambda)  (None, 896, 3072)    0           tf.math.multiply_14[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "dropout2_celltype (Dropout)     (None, 3072)         0           dense_combn2_celltype[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_5 (TFOpLambda)  (None, 896, 3072)    0           tf.math.multiply_10[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_15 (TFOpLambda (None, 896, 3072)    0           tf.math.sigmoid_7[0][0]          \n",
      "                                                                 dropout1_finalpoint_quant[0][0]  \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_1 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_2 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_3 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_4 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_5 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_chan_celltype_6 (Dense)   (None, 896)          2753408     dropout2_celltype[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_11 (TFOpLambda (None, 896, 3072)    0           tf.math.sigmoid_5[0][0]          \n",
      "                                                                 dropout1_finalpoint[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_1 (Dense)        (None, 896, 10)      30730       tf.math.multiply_15[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_2 (Dense)        (None, 896, 10)      30730       tf.math.multiply_15[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_3 (Dense)        (None, 896, 10)      30730       tf.math.multiply_15[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_4 (Dense)        (None, 896, 10)      30730       tf.math.multiply_15[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_5 (Dense)        (None, 896, 10)      30730       tf.math.multiply_15[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "quant_out_chan_6 (Dense)        (None, 896, 10)      30730       tf.math.multiply_15[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_12 (TFOpLambda)  (None, 896, 1)       0           dense_chan_celltype_1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_13 (TFOpLambda)  (None, 896, 1)       0           dense_chan_celltype_2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_14 (TFOpLambda)  (None, 896, 1)       0           dense_chan_celltype_3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_15 (TFOpLambda)  (None, 896, 1)       0           dense_chan_celltype_4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_16 (TFOpLambda)  (None, 896, 1)       0           dense_chan_celltype_5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_17 (TFOpLambda)  (None, 896, 1)       0           dense_chan_celltype_6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "avg (Dense)                     (None, 896, 6)       18438       tf.math.multiply_11[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_18 (TFOpLambda)  (None, 896, 1, 10)   0           quant_out_chan_1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_19 (TFOpLambda)  (None, 896, 1, 10)   0           quant_out_chan_2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_20 (TFOpLambda)  (None, 896, 1, 10)   0           quant_out_chan_3[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_21 (TFOpLambda)  (None, 896, 1, 10)   0           quant_out_chan_4[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_22 (TFOpLambda)  (None, 896, 1, 10)   0           quant_out_chan_5[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_23 (TFOpLambda)  (None, 896, 1, 10)   0           quant_out_chan_6[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "delta (Concatenate)             (None, 896, 6)       0           tf.expand_dims_12[0][0]          \n",
      "                                                                 tf.expand_dims_13[0][0]          \n",
      "                                                                 tf.expand_dims_14[0][0]          \n",
      "                                                                 tf.expand_dims_15[0][0]          \n",
      "                                                                 tf.expand_dims_16[0][0]          \n",
      "                                                                 tf.expand_dims_17[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_24 (TFOpLambda)  (None, 896, 6, 1)    0           avg[0][0]                        \n",
      "__________________________________________________________________________________________________\n",
      "quant (Concatenate)             (None, 896, 6, 10)   0           tf.expand_dims_18[0][0]          \n",
      "                                                                 tf.expand_dims_19[0][0]          \n",
      "                                                                 tf.expand_dims_20[0][0]          \n",
      "                                                                 tf.expand_dims_21[0][0]          \n",
      "                                                                 tf.expand_dims_22[0][0]          \n",
      "                                                                 tf.expand_dims_23[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_25 (TFOpLambda)  (None, 896, 6, 1)    0           delta[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "cat_avg_delta (Concatenate)     (None, 896, 6, 12)   0           tf.expand_dims_24[0][0]          \n",
      "                                                                 quant[0][0]                      \n",
      "                                                                 tf.expand_dims_25[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.__operators__.getitem (Slici (None, 896, 12)      0           cat_avg_delta[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "tf.__operators__.getitem_1 (Sli (None, 896, 12)      0           cat_avg_delta[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "tf.__operators__.getitem_2 (Sli (None, 896, 12)      0           cat_avg_delta[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "tf.__operators__.getitem_3 (Sli (None, 896, 12)      0           cat_avg_delta[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "tf.__operators__.getitem_4 (Sli (None, 896, 12)      0           cat_avg_delta[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "tf.__operators__.getitem_5 (Sli (None, 896, 12)      0           cat_avg_delta[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_chan_0 (BatchNormali (None, 896, 12)      48          tf.__operators__.getitem[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_chan_1 (BatchNormali (None, 896, 12)      48          tf.__operators__.getitem_1[0][0] \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_chan_2 (BatchNormali (None, 896, 12)      48          tf.__operators__.getitem_2[0][0] \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_chan_3 (BatchNormali (None, 896, 12)      48          tf.__operators__.getitem_3[0][0] \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_chan_4 (BatchNormali (None, 896, 12)      48          tf.__operators__.getitem_4[0][0] \n",
      "__________________________________________________________________________________________________\n",
      "batchnorm1_chan_5 (BatchNormali (None, 896, 12)      48          tf.__operators__.getitem_5[0][0] \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_16 (TFOpLambda (None, 896, 12)      0           batchnorm1_chan_0[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_20 (TFOpLambda (None, 896, 12)      0           batchnorm1_chan_1[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_24 (TFOpLambda (None, 896, 12)      0           batchnorm1_chan_2[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_28 (TFOpLambda (None, 896, 12)      0           batchnorm1_chan_3[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_32 (TFOpLambda (None, 896, 12)      0           batchnorm1_chan_4[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_36 (TFOpLambda (None, 896, 12)      0           batchnorm1_chan_5[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_8 (TFOpLambda)  (None, 896, 12)      0           tf.math.multiply_16[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_10 (TFOpLambda) (None, 896, 12)      0           tf.math.multiply_20[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_12 (TFOpLambda) (None, 896, 12)      0           tf.math.multiply_24[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_14 (TFOpLambda) (None, 896, 12)      0           tf.math.multiply_28[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_16 (TFOpLambda) (None, 896, 12)      0           tf.math.multiply_32[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_18 (TFOpLambda) (None, 896, 12)      0           tf.math.multiply_36[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_17 (TFOpLambda (None, 896, 12)      0           tf.math.sigmoid_8[0][0]          \n",
      "                                                                 batchnorm1_chan_0[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_21 (TFOpLambda (None, 896, 12)      0           tf.math.sigmoid_10[0][0]         \n",
      "                                                                 batchnorm1_chan_1[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_25 (TFOpLambda (None, 896, 12)      0           tf.math.sigmoid_12[0][0]         \n",
      "                                                                 batchnorm1_chan_2[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_29 (TFOpLambda (None, 896, 12)      0           tf.math.sigmoid_14[0][0]         \n",
      "                                                                 batchnorm1_chan_3[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_33 (TFOpLambda (None, 896, 12)      0           tf.math.sigmoid_16[0][0]         \n",
      "                                                                 batchnorm1_chan_4[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_37 (TFOpLambda (None, 896, 12)      0           tf.math.sigmoid_18[0][0]         \n",
      "                                                                 batchnorm1_chan_5[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_chan_0 (Conv1D)         (None, 896, 96)      1248        tf.math.multiply_17[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_chan_1 (Conv1D)         (None, 896, 96)      1248        tf.math.multiply_21[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_chan_2 (Conv1D)         (None, 896, 96)      1248        tf.math.multiply_25[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_chan_3 (Conv1D)         (None, 896, 96)      1248        tf.math.multiply_29[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_chan_4 (Conv1D)         (None, 896, 96)      1248        tf.math.multiply_33[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv1d1_chan_5 (Conv1D)         (None, 896, 96)      1248        tf.math.multiply_37[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_chan_0 (Dropout)       (None, 896, 96)      0           conv1d1_chan_0[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_chan_1 (Dropout)       (None, 896, 96)      0           conv1d1_chan_1[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_chan_2 (Dropout)       (None, 896, 96)      0           conv1d1_chan_2[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_chan_3 (Dropout)       (None, 896, 96)      0           conv1d1_chan_3[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_chan_4 (Dropout)       (None, 896, 96)      0           conv1d1_chan_4[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dropout1_chan_5 (Dropout)       (None, 896, 96)      0           conv1d1_chan_5[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_18 (TFOpLambda (None, 896, 96)      0           dropout1_chan_0[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_22 (TFOpLambda (None, 896, 96)      0           dropout1_chan_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_26 (TFOpLambda (None, 896, 96)      0           dropout1_chan_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_30 (TFOpLambda (None, 896, 96)      0           dropout1_chan_3[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_34 (TFOpLambda (None, 896, 96)      0           dropout1_chan_4[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_38 (TFOpLambda (None, 896, 96)      0           dropout1_chan_5[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_9 (TFOpLambda)  (None, 896, 96)      0           tf.math.multiply_18[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_11 (TFOpLambda) (None, 896, 96)      0           tf.math.multiply_22[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_13 (TFOpLambda) (None, 896, 96)      0           tf.math.multiply_26[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_15 (TFOpLambda) (None, 896, 96)      0           tf.math.multiply_30[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_17 (TFOpLambda) (None, 896, 96)      0           tf.math.multiply_34[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.sigmoid_19 (TFOpLambda) (None, 896, 96)      0           tf.math.multiply_38[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_19 (TFOpLambda (None, 896, 96)      0           tf.math.sigmoid_9[0][0]          \n",
      "                                                                 dropout1_chan_0[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_23 (TFOpLambda (None, 896, 96)      0           tf.math.sigmoid_11[0][0]         \n",
      "                                                                 dropout1_chan_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_27 (TFOpLambda (None, 896, 96)      0           tf.math.sigmoid_13[0][0]         \n",
      "                                                                 dropout1_chan_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_31 (TFOpLambda (None, 896, 96)      0           tf.math.sigmoid_15[0][0]         \n",
      "                                                                 dropout1_chan_3[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_35 (TFOpLambda (None, 896, 96)      0           tf.math.sigmoid_17[0][0]         \n",
      "                                                                 dropout1_chan_4[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "tf.math.multiply_39 (TFOpLambda (None, 896, 96)      0           tf.math.sigmoid_19[0][0]         \n",
      "                                                                 dropout1_chan_5[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "flatten_chan_0 (Flatten)        (None, 86016)        0           tf.math.multiply_19[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "flatten_chan_1 (Flatten)        (None, 86016)        0           tf.math.multiply_23[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "flatten_chan_2 (Flatten)        (None, 86016)        0           tf.math.multiply_27[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "flatten_chan_3 (Flatten)        (None, 86016)        0           tf.math.multiply_31[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "flatten_chan_4 (Flatten)        (None, 86016)        0           tf.math.multiply_35[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "flatten_chan_5 (Flatten)        (None, 86016)        0           tf.math.multiply_39[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "avg_delta_out_channel0 (Dense)  (None, 896)          77071232    flatten_chan_0[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "avg_delta_out_channel1 (Dense)  (None, 896)          77071232    flatten_chan_1[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "avg_delta_out_channel2 (Dense)  (None, 896)          77071232    flatten_chan_2[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "avg_delta_out_channel3 (Dense)  (None, 896)          77071232    flatten_chan_3[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "avg_delta_out_channel4 (Dense)  (None, 896)          77071232    flatten_chan_4[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "avg_delta_out_channel5 (Dense)  (None, 896)          77071232    flatten_chan_5[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_26 (TFOpLambda)  (None, 896, 1)       0           avg_delta_out_channel0[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_27 (TFOpLambda)  (None, 896, 1)       0           avg_delta_out_channel1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_28 (TFOpLambda)  (None, 896, 1)       0           avg_delta_out_channel2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_29 (TFOpLambda)  (None, 896, 1)       0           avg_delta_out_channel3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_30 (TFOpLambda)  (None, 896, 1)       0           avg_delta_out_channel4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "tf.expand_dims_31 (TFOpLambda)  (None, 896, 1)       0           avg_delta_out_channel5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "cat_combn_avg_delta (Concatenat (None, 896, 6)       0           tf.expand_dims_26[0][0]          \n",
      "                                                                 tf.expand_dims_27[0][0]          \n",
      "                                                                 tf.expand_dims_28[0][0]          \n",
      "                                                                 tf.expand_dims_29[0][0]          \n",
      "                                                                 tf.expand_dims_30[0][0]          \n",
      "                                                                 tf.expand_dims_31[0][0]          \n",
      "==================================================================================================\n",
      "Total params: 2,151,696,972\n",
      "Trainable params: 2,151,690,684\n",
      "Non-trainable params: 6,288\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = Enformer_Celltyping(use_prebuilt_model=True,\n",
    "                            #path to stage 1 (split) training weights\n",
    "                            enf_celltyping_pth = str(PROJECT_PATH / 'test_split_weights.h5')\n",
    "                           )\n",
    "\n",
    "learning_rate = 0.0005 #higher learning rate than pretraining\n",
    "#compile\n",
    "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=learning_rate),\n",
    "              loss=tf.keras.losses.poisson,\n",
    "              metrics=['mse',pearsonR])\n",
    "#Let's view the model\n",
    "model.summary() #note more trainable parameters"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c1527d98",
   "metadata": {},
   "source": [
    "Like with pretraining, this model does not include the layers of enformer since the DNA for the demo regions has already been passed through the pre-trained enformer model. Since the layers are frozen these weights will not update so we can save on compute time and RAM by pre-running it. If you want to build a version of Enformer Celltyping with the enformer layers run:\n",
    "\n",
    "```\n",
    "from EnformerCelltyping.constants import DATA_PATH\n",
    "from EnformerCelltyping.enf_celltyping import Enformer_Celltyping\n",
    "\n",
    "assays = ['h3k27ac', 'h3k4me1', 'h3k4me3', 'h3k9me3', 'h3k27me3', 'h3k36me3']\n",
    "\n",
    "#Enformer Celltyping from scratch\n",
    "#Include enformer layers\n",
    "model = Enformer_Celltyping(assays=assays,\n",
    "                            use_prebuilt_model=True,\n",
    "                            enf_celltyping_pth = str(PROJECT_PATH / 'test_split_weights.h5'),\n",
    "                            enf_path = str(DATA_PATH / \"enformer_model\"))\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3bba5edf",
   "metadata": {},
   "source": [
    "Next, we need to update our dataloader to give the cell type-specific outputs not separate channels:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "85d2d27f",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Now use the provided custom data loader on this data\n",
    "from EnformerCelltyping.utils import PreSavedDataGen\n",
    "\n",
    "train_dataloader = PreSavedDataGen(files=train_data,\n",
    "                                   batch_size=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "baae1c0a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X input to model shape\n",
      "{'dna': TensorShape([1, 896, 1536]), 'chrom_access_lcl': TensorShape([1, 1562]), 'chrom_access_gbl': TensorShape([1, 14592])}\n",
      "Y output from model shape\n",
      "{'y': TensorShape([1, 896, 6])}\n"
     ]
    }
   ],
   "source": [
    "#test ouput\n",
    "one_samp = next(iter(train_dataloader))\n",
    "#shape of X inputs\n",
    "print(\"X input to model shape\")\n",
    "print({key: dat.shape for key, dat in one_samp[0].items()})\n",
    "#shape of Y inputs\n",
    "print(\"Y output from model shape\")\n",
    "print({'y': one_samp[1].shape})\n",
    "#You can see there is just one output now for y - the cell type-specific signal"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "be627d8e",
   "metadata": {},
   "source": [
    "\n",
    "\n",
    "Now let's train Enformer Celltyping"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "284d8289",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "1/1 - 9s - loss: 0.8667 - mse: 0.2238 - pearsonR: -5.1377e-03\n",
      "Epoch 2/20\n",
      "1/1 - 5s - loss: 0.8830 - mse: 0.4591 - pearsonR: 0.0376\n",
      "Epoch 3/20\n",
      "1/1 - 5s - loss: 0.7279 - mse: 0.0683 - pearsonR: 0.8450\n",
      "Epoch 4/20\n",
      "1/1 - 5s - loss: 0.8512 - mse: 0.4223 - pearsonR: 0.0269\n",
      "Epoch 5/20\n",
      "1/1 - 5s - loss: 0.8179 - mse: 0.2546 - pearsonR: 0.0680\n",
      "Epoch 6/20\n",
      "1/1 - 5s - loss: 0.7064 - mse: 0.2834 - pearsonR: 0.8185\n",
      "Epoch 7/20\n",
      "1/1 - 5s - loss: 0.6195 - mse: 0.0646 - pearsonR: 0.9235\n",
      "Epoch 8/20\n",
      "1/1 - 5s - loss: 0.7700 - mse: 0.1369 - pearsonR: 0.0014\n",
      "Epoch 9/20\n",
      "1/1 - 5s - loss: 0.6713 - mse: 0.1871 - pearsonR: 0.8188\n",
      "Epoch 10/20\n",
      "1/1 - 5s - loss: 0.8592 - mse: 0.3274 - pearsonR: 0.0751\n",
      "Epoch 11/20\n",
      "1/1 - 5s - loss: 0.7585 - mse: 0.1538 - pearsonR: 0.8483\n",
      "Epoch 12/20\n",
      "1/1 - 5s - loss: 0.7601 - mse: 0.1991 - pearsonR: 0.7637\n",
      "Epoch 13/20\n",
      "1/1 - 5s - loss: 1.0428 - mse: 0.4666 - pearsonR: 0.0634\n",
      "Epoch 14/20\n",
      "1/1 - 5s - loss: 0.9290 - mse: 0.2159 - pearsonR: 0.0326\n",
      "Epoch 15/20\n",
      "1/1 - 5s - loss: 0.8364 - mse: 0.2649 - pearsonR: 0.4053\n",
      "Epoch 16/20\n",
      "1/1 - 5s - loss: 0.7123 - mse: 0.1137 - pearsonR: 0.6902\n",
      "Epoch 17/20\n",
      "1/1 - 5s - loss: 0.7295 - mse: 0.2522 - pearsonR: 0.6051\n",
      "Epoch 18/20\n",
      "1/1 - 5s - loss: 0.8188 - mse: 0.2953 - pearsonR: 0.8004\n",
      "Epoch 19/20\n",
      "1/1 - 5s - loss: 0.5862 - mse: 0.0357 - pearsonR: 0.9530\n",
      "Epoch 20/20\n",
      "1/1 - 5s - loss: 0.8636 - mse: 0.2602 - pearsonR: 0.7241\n",
      "Time taken:  0:01:42.903473\n"
     ]
    }
   ],
   "source": [
    "n_steps = 20\n",
    "\n",
    "import datetime\n",
    "strt = datetime.datetime.now()\n",
    "\n",
    "# Train the model\n",
    "model.fit(\n",
    "        train_dataloader,\n",
    "        epochs=n_steps,\n",
    "        verbose=2\n",
    "    )\n",
    "\n",
    "\n",
    "end = datetime.datetime.now()\n",
    "print('Time taken: ',end-strt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "53f29374",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1200x600 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Now let's predict one of the regions\n",
    "#function to plot multiple tracks with the mdoel prediction, the actual and the local chrom access\n",
    "\n",
    "#actual signal\n",
    "signal = one_samp[1].numpy()\n",
    "\n",
    "#We can get the local chromatin accessibility data from X\n",
    "#This is the change in chromatin acessibility between the\n",
    "#cell type of interest and the avg from trainig cell types.\n",
    "chrom_access_lcl = one_samp[0]['chrom_access_lcl'].numpy()\n",
    "#chrom access local is wider than output window so need to cut it\n",
    "buff = (chrom_access_lcl.shape[1] - signal.shape[1])//2\n",
    "chrom_access_lcl = chrom_access_lcl[:,buff:chrom_access_lcl.shape[1]-buff]\n",
    "#now just repeat track for each plot and swap axis\n",
    "import numpy as np\n",
    "chrom_access_lcl = np.tile(chrom_access_lcl,[6,1]).swapaxes(0, 1)\n",
    "\n",
    "\n",
    "model_pred = model.predict(one_samp[0])\n",
    "\n",
    "from EnformerCelltyping.utils import plot_tracks\n",
    "#just plot h3k27ac\n",
    "plot_tracks(signal[0,:,0:1],model_pred[0,:,0:1],chrom_access_lcl[:,0:1],\n",
    "            labels=['h3k27ac'],\n",
    "            same_y=False,\n",
    "            nme_p1='Experiment',nme_p2='Enformer Celltyping',\n",
    "            nme_p3='Chromatin Accessibility Delta',\n",
    "            pal =[\"#9A8822\",\"#F8AFA8\",\"#74A089\"],\n",
    "            figsize=(12., 6.))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d5e3d5c0",
   "metadata": {},
   "source": [
    "Note that performance here is very poor so we need to train longer, on more data and avoid overfitting! But I hope this explains the training process"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "EnformerCelltyping",
   "language": "python",
   "name": "enformercelltyping"
  },
  "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
}