2224 lines (2223 with data), 78.0 kB
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"This is a temporary notebook for shap value analysis and plots."
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import sys\n",
"\n",
"proj_dir = \"/home/scai/PhenPred\"\n",
"if not os.path.exists(proj_dir):\n",
" proj_dir = \"/Users/emanuel/Projects/PhenPred\"\n",
"sys.path.extend([proj_dir])\n",
"\n",
"import json\n",
"import PhenPred\n",
"import argparse\n",
"import pandas as pd\n",
"from PhenPred.vae import plot_folder\n",
"from PhenPred.vae.Hypers import Hypers\n",
"from PhenPred.vae.Train import CLinesTrain\n",
"from PhenPred.vae.DatasetDepMap23Q2 import CLinesDatasetDepMap23Q2\n",
"from PhenPred.vae.DatasetMOFA import CLinesDatasetMOFA\n",
"from PhenPred.vae.DatasetMOVE import CLinesDatasetMOVE\n",
"from PhenPred.vae.DatasetJAMIE import CLinesDatasetJAMIE\n",
"from PhenPred.vae.DatasetIClusterPlus import CLinesDatasetIClusterPlus\n",
"from PhenPred.vae.DatasetMoCluster import CLinesDatasetMoCluster\n",
"from PhenPred.vae.DatasetMixOmics import CLinesDatasetMixOmics\n",
"from PhenPred.vae.DatasetSCVAEIT import CLinesDatasetSCVAEIT\n",
"from PhenPred.Utils import two_vars_correlation\n",
"from sklearn.discriminant_analysis import StandardScaler\n",
"from sklearn.metrics import calinski_harabasz_score, davies_bouldin_score\n",
"\n",
"\n",
"pd.set_option(\"display.max_rows\", 100)\n",
"pd.set_option(\"display.max_columns\", 100)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"from scipy.stats import ttest_ind, ttest_rel, wilcoxon\n",
"from scipy.stats import shapiro"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"import warnings\n",
"\n",
"warnings.filterwarnings(\"ignore\")\n",
"import matplotlib.pyplot as plt\n",
"import numpy as np\n",
"import pandas as pd\n",
"import seaborn as sns\n",
"\n",
"plt.rcParams[\"font.family\"] = \"Arial\"\n",
"plt.rcParams[\"font.size\"] = 4\n",
"plt.rcParams[\"axes.linewidth\"] = 0.25\n",
"plt.rcParams[\"figure.figsize\"] = (2.5, 2.5)\n",
"plt.rcParams[\"pdf.fonttype\"] = 42\n",
"plt.rcParams[\"ps.fonttype\"] = 42\n",
"plt.rcParams[\"figure.dpi\"] = 200\n",
"plt.rcParams[\"axes.linewidth\"] = 0.25\n",
"plt.rcParams[\"legend.fontsize\"] = 4\n",
"\n",
"sns.set(style=\"ticks\", context=\"paper\", font_scale=1, font=\"Arial\")\n",
"sns.set_context(\n",
" \"paper\",\n",
" rc={\n",
" \"axes.linewidth\": 0.25,\n",
" \"xtick.major.size\": 2,\n",
" \"xtick.major.width\": 0.25,\n",
" \"ytick.major.size\": 2,\n",
" \"ytick.major.width\": 0.25,\n",
" \"xtick.labelsize\": 6,\n",
" \"ytick.labelsize\": 6,\n",
" \"axes.labelsize\": 7,\n",
" \"legend.fontsize\": 6,\n",
" \"legend.title_fontsize\": 6,\n",
" },\n",
")\n",
"\n",
"import matplotlib.patches as mpatches\n",
"import umap\n",
"\n",
"pd.set_option(\"display.max_rows\", 100)\n",
"pd.set_option(\"display.max_columns\", 100)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"import shap\n",
"import pickle\n",
"from tqdm.notebook import tqdm"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"OMIC_PALLETS = {\n",
" \"conditionals\": \"#4c72b0\",\n",
" \"copynumber\": \"#dd8452\",\n",
" \"drugresponse\": \"#55a868\",\n",
" \"metabolomics\": \"#c44e52\",\n",
" \"proteomics\": \"#8172b3\",\n",
" \"crisprcas9\": \"#937860\",\n",
" \"transcriptomics\": \"#da8bc3\",\n",
" \"methylation\": \"#8c8c8c\",\n",
"}"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Latent comparison"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [],
"source": [
"TIMESTAMP = \"20240830_110319\""
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [],
"source": [
"clustering_score_df = pd.read_csv(\n",
" f\"./reports/vae/latent/{TIMESTAMP}_clustering_score.csv\"\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [],
"source": [
"_, ax = plt.subplots(1, 1, figsize=(3, 3), dpi=600)\n",
"sns.barplot(data=clustering_score_df, x=\"metric\", y=\"score\", ax=ax, hue=\"model\")\n",
"PhenPred.save_figure(f\"{plot_folder}/latent/{TIMESTAMP}_clustering_score_barplot_2vs7omcis\")"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"TIMESTAMP_2omics = \"20240830_110319\"\n",
"TIMESTAMP_7omics = \"20231023_092657\""
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"<not serializable>\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"copynumber\": \"data/clines//cnv_summary_20230303_matrix.csv\",\n",
" \"crisprcas9\": \"data/clines//depmap23Q2/CRISPRGeneEffect.csv\",\n",
" \"drugresponse\": \"data/clines//drugresponse.csv\",\n",
" \"metabolomics\": \"data/clines//metabolomics.csv\",\n",
" \"methylation\": \"data/clines//methylation.csv\",\n",
" \"proteomics\": \"data/clines//proteomics.csv\",\n",
" \"transcriptomics\": \"data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": \"20231023_092657\",\n",
" \"model\": \"MOVE\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"<not serializable>\",\n",
" \"save_model\": true,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_cv\": false,\n",
" \"standardize\": true,\n",
" \"use_conditionals\": true,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\",\n",
" \"mean\",\n",
" \"mean\",\n",
" \"mean\",\n",
" \"mean\",\n",
" \"macro\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"DepMap23Q2 | Samples = 1,656 | Proteomics = 4,922 (0 masked) | Metabolomics = 225 (0 masked) | Drug response = 810 (0 masked) | CRISPR-Cas9 = 17,931 (12,714 masked) | Methylation = 14,608 (7,014 masked) | Transcriptomics = 15,278 (7,193 masked) | Copy number = 777 (0 masked) | Labels = 237\n"
]
}
],
"source": [
"hyperparameters_7omics = Hypers.read_hyperparameters(timestamp=\"20231023_092657\")\n",
"clines_db_7omics = CLinesDatasetDepMap23Q2(\n",
" datasets=hyperparameters_7omics[\"datasets\"],\n",
" labels_names=hyperparameters_7omics[\"labels\"],\n",
" standardize=hyperparameters_7omics[\"standardize\"],\n",
" filter_features=hyperparameters_7omics[\"filter_features\"],\n",
" filtered_encoder_only=hyperparameters_7omics[\"filtered_encoder_only\"],\n",
" feature_miss_rate_thres=hyperparameters_7omics[\"feature_miss_rate_thres\"],\n",
")\n",
"\n",
"train_7omics = CLinesTrain(\n",
" clines_db_7omics,\n",
" hyperparameters_7omics,\n",
" verbose=hyperparameters_7omics[\"verbose\"],\n",
" stratify_cv_by=clines_db_7omics.samples_by_tissue(\n",
" \"Haematopoietic and Lymphoid\"\n",
" ),\n",
")\n",
"train_7omics.run(run_timestamp=hyperparameters_7omics[\"load_run\"])\n",
"mosa_7omics_imputed, mosa_7omics_latent = train_7omics.load_vae_reconstructions()\n",
"mosa_7omics_predicted, _ = train_7omics.load_vae_reconstructions(mode=\"all\")"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"<not serializable>\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": \"20240830_110319\",\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"<not serializable>\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": true,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"DepMap23Q2 | Samples = 1,590 | Transcriptomics = 15,278 (7,193 masked) | Drug response = 810 (0 masked) | Labels = 237\n"
]
}
],
"source": [
"hyperparameters = Hypers.read_hyperparameters(timestamp=TIMESTAMP_2omics)\n",
"clines_db = CLinesDatasetDepMap23Q2(\n",
" labels_names=hyperparameters[\"labels\"],\n",
" datasets=hyperparameters[\"datasets\"],\n",
" feature_miss_rate_thres=hyperparameters[\"feature_miss_rate_thres\"],\n",
" standardize=hyperparameters[\"standardize\"],\n",
" filter_features=hyperparameters[\"filter_features\"],\n",
" filtered_encoder_only=hyperparameters[\"filtered_encoder_only\"],\n",
")\n",
"train = CLinesTrain(\n",
" clines_db,\n",
" hyperparameters,\n",
" verbose=hyperparameters[\"verbose\"],\n",
" stratify_cv_by=clines_db.samples_by_tissue(\"Haematopoietic and Lymphoid\"),\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n"
]
}
],
"source": [
"train.run(run_timestamp=hyperparameters[\"load_run\"])\n",
"vae_imputed, vae_latent = train.load_vae_reconstructions()\n",
"vae_predicted, _ = train.load_vae_reconstructions(mode=\"all\")\n",
"\n",
"mofa_imputed, mofa_latent = CLinesDatasetMOFA.load_reconstructions(clines_db)\n",
"move_diabetes_imputed, move_diabetes_latent = CLinesDatasetMOVE.load_reconstructions(\n",
" clines_db\n",
")\n",
"jamie_imputed, jamie_latent = CLinesDatasetJAMIE.load_reconstructions(clines_db)\n",
"\n",
"_, mixOmics_latent = CLinesDatasetMixOmics.load_reconstructions(clines_db)\n",
"_, iClusterPlus_latent = CLinesDatasetIClusterPlus.load_reconstructions(clines_db)\n",
"_, moCluster_latent = CLinesDatasetMoCluster.load_reconstructions(clines_db)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [],
"source": [
"common_cell_lines = list(\n",
" set(vae_latent.index)\n",
" & set(mosa_7omics_latent.index)\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [],
"source": [
"mosa_7omics_latent = mosa_7omics_latent.loc[common_cell_lines]\n",
"vae_latent = vae_latent.loc[common_cell_lines]\n",
"mofa_latent[\"factors\"] = mofa_latent[\"factors\"].loc[common_cell_lines]\n",
"move_diabetes_latent[\"factors\"] = move_diabetes_latent[\"factors\"].loc[common_cell_lines]\n",
"jamie_latent[\"factors\"] = jamie_latent[\"factors\"].loc[common_cell_lines]\n",
"mixOmics_latent[\"factors\"] = mixOmics_latent[\"factors\"].loc[common_cell_lines]\n",
"iClusterPlus_latent[\"factors\"] = iClusterPlus_latent[\"factors\"].loc[common_cell_lines]\n",
"moCluster_latent[\"factors\"] = moCluster_latent[\"factors\"].loc[common_cell_lines]"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [],
"source": [
"samplesheet = clines_db.samplesheet[\"tissue\"].fillna(\"Other tissue\")"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [],
"source": [
"clustering_score_df = {\"model\": [], \"metric\": [], \"score\": []}\n",
"for n, z_joint in [\n",
" (\"MOSA (7 omics)\", mosa_7omics_latent),\n",
" (\"MOSA (2 omics)\", vae_latent),\n",
" (\"MOFA\", mofa_latent[\"factors\"]),\n",
" (\"MOVE\", move_diabetes_latent[\"factors\"]),\n",
" (\"JAMIE\", jamie_latent[\"factors\"]),\n",
" (\"mixOmics\", mixOmics_latent[\"factors\"]),\n",
" (\"iClusterPlus\", iClusterPlus_latent[\"factors\"]),\n",
" (\"moCluster\", moCluster_latent[\"factors\"]),\n",
"]:\n",
" cluster_labels = samplesheet[z_joint.index]\n",
" clustering_score_df[\"model\"].append(n)\n",
" clustering_score_df[\"metric\"].append(\"calinski_harabasz\")\n",
" clustering_score_df[\"score\"].append(\n",
" calinski_harabasz_score(\n",
" StandardScaler().fit_transform(z_joint), cluster_labels\n",
" )\n",
" )\n",
" clustering_score_df[\"model\"].append(n)\n",
" clustering_score_df[\"metric\"].append(\"davies_bouldin\")\n",
" clustering_score_df[\"score\"].append(\n",
" davies_bouldin_score(\n",
" StandardScaler().fit_transform(z_joint), cluster_labels\n",
" )\n",
" )\n",
"clustering_score_df = pd.DataFrame(clustering_score_df)"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [],
"source": [
"_, ax = plt.subplots(1, 1, figsize=(3, 3), dpi=600)\n",
"sns.barplot(data=clustering_score_df, x=\"metric\", y=\"score\", ax=ax, hue=\"model\")\n",
"PhenPred.save_figure(\n",
" f\"{plot_folder}/latent/{TIMESTAMP_2omics}_clustering_score_barplot_2vs7omics\"\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# RNA"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [],
"source": [
"TIMESTAMP_2omics = \"20240830_110319\"\n",
"TIMESTAMP_7omics = \"20231023_092657\""
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"<not serializable>\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"copynumber\": \"data/clines//cnv_summary_20230303_matrix.csv\",\n",
" \"crisprcas9\": \"data/clines//depmap23Q2/CRISPRGeneEffect.csv\",\n",
" \"drugresponse\": \"data/clines//drugresponse.csv\",\n",
" \"metabolomics\": \"data/clines//metabolomics.csv\",\n",
" \"methylation\": \"data/clines//methylation.csv\",\n",
" \"proteomics\": \"data/clines//proteomics.csv\",\n",
" \"transcriptomics\": \"data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": \"20231023_092657\",\n",
" \"model\": \"MOVE\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"<not serializable>\",\n",
" \"save_model\": true,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_cv\": false,\n",
" \"standardize\": true,\n",
" \"use_conditionals\": true,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\",\n",
" \"mean\",\n",
" \"mean\",\n",
" \"mean\",\n",
" \"mean\",\n",
" \"macro\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"DepMap23Q2 | Samples = 1,656 | Proteomics = 4,922 (0 masked) | Metabolomics = 225 (0 masked) | Drug response = 810 (0 masked) | CRISPR-Cas9 = 17,931 (12,714 masked) | Methylation = 14,608 (7,014 masked) | Transcriptomics = 15,278 (7,193 masked) | Copy number = 777 (0 masked) | Labels = 237\n"
]
}
],
"source": [
"hyperparameters = Hypers.read_hyperparameters(timestamp=TIMESTAMP_7omics)\n",
"clines_db = CLinesDatasetDepMap23Q2(\n",
" labels_names=hyperparameters[\"labels\"],\n",
" datasets=hyperparameters[\"datasets\"],\n",
" feature_miss_rate_thres=hyperparameters[\"feature_miss_rate_thres\"],\n",
" standardize=hyperparameters[\"standardize\"],\n",
" filter_features=hyperparameters[\"filter_features\"],\n",
" filtered_encoder_only=hyperparameters[\"filtered_encoder_only\"],\n",
")\n",
"train = CLinesTrain(\n",
" clines_db,\n",
" hyperparameters,\n",
" verbose=hyperparameters[\"verbose\"],\n",
" stratify_cv_by=clines_db.samples_by_tissue(\"Haematopoietic and Lymphoid\"),\n",
")\n",
"\n",
"train.run(run_timestamp=hyperparameters[\"load_run\"])\n",
"mosa_7omics_imputed, mosa_7omics_latent = train.load_vae_reconstructions()\n",
"mosa_7omics_predicted, _ = train.load_vae_reconstructions(mode=\"all\")"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"<not serializable>\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": \"20240830_110319\",\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"<not serializable>\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": true,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"DepMap23Q2 | Samples = 1,590 | Transcriptomics = 15,278 (7,193 masked) | Drug response = 810 (0 masked) | Labels = 237\n"
]
}
],
"source": [
"hyperparameters = Hypers.read_hyperparameters(timestamp=TIMESTAMP_2omics)\n",
"clines_db = CLinesDatasetDepMap23Q2(\n",
" labels_names=hyperparameters[\"labels\"],\n",
" datasets=hyperparameters[\"datasets\"],\n",
" feature_miss_rate_thres=hyperparameters[\"feature_miss_rate_thres\"],\n",
" standardize=hyperparameters[\"standardize\"],\n",
" filter_features=hyperparameters[\"filter_features\"],\n",
" filtered_encoder_only=hyperparameters[\"filtered_encoder_only\"],\n",
")\n",
"train = CLinesTrain(\n",
" clines_db,\n",
" hyperparameters,\n",
" verbose=hyperparameters[\"verbose\"],\n",
" stratify_cv_by=clines_db.samples_by_tissue(\"Haematopoietic and Lymphoid\"),\n",
")\n",
"\n",
"train.run(run_timestamp=hyperparameters[\"load_run\"])"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n",
"# ---- Hyperparameters\n",
"{\n",
" \"activation_function\": \"prelu\",\n",
" \"batch_norm\": false,\n",
" \"batch_size\": 256,\n",
" \"contrastive_neg_margin\": 0.15,\n",
" \"contrastive_pos_margin\": 0.85,\n",
" \"dataname\": \"depmap23Q2\",\n",
" \"datasets\": {\n",
" \"drugresponse\": \"/home/scai/PhenPred/data/clines//drugresponse.csv\",\n",
" \"transcriptomics\": \"/home/scai/PhenPred/data/clines//depmap23Q2/OmicsExpressionGenesExpectedCountProfileVoom.csv\"\n",
" },\n",
" \"feature_dropout\": 0,\n",
" \"feature_miss_rate_thres\": 0.85,\n",
" \"filter_features\": [\n",
" \"transcriptomics\",\n",
" \"crisprcas9\",\n",
" \"methylation\"\n",
" ],\n",
" \"filtered_encoder_only\": true,\n",
" \"gmvae_decay_temp\": true,\n",
" \"gmvae_decay_temp_rate\": 0.013862944,\n",
" \"gmvae_hard_gumbel\": 0.7936881144482251,\n",
" \"gmvae_hidden_size\": 935,\n",
" \"gmvae_init_temp\": 1.0,\n",
" \"gmvae_k\": 51,\n",
" \"gmvae_min_temp\": 0.5,\n",
" \"gmvae_views_logits\": 726,\n",
" \"hidden_dims\": [\n",
" 0.7\n",
" ],\n",
" \"labels\": [\n",
" \"tissue\",\n",
" \"mutations\",\n",
" \"fussions\",\n",
" \"msi\",\n",
" \"growth\"\n",
" ],\n",
" \"latent_dim\": 200,\n",
" \"learning_rate\": 0.0003,\n",
" \"load_run\": null,\n",
" \"model\": \"MOSA\",\n",
" \"n_folds\": 3,\n",
" \"num_epochs\": 500,\n",
" \"optimizer_type\": \"adam\",\n",
" \"probability\": 0.4,\n",
" \"reconstruction_loss\": \"mse\",\n",
" \"save_model\": false,\n",
" \"scheduler\": \"plateau\",\n",
" \"scheduler_factor\": 0.6,\n",
" \"scheduler_min_lr\": 1e-07,\n",
" \"scheduler_patience\": 7,\n",
" \"scheduler_threshold\": 0.0001,\n",
" \"skip_benchmarks\": false,\n",
" \"skip_cv\": true,\n",
" \"standardize\": true,\n",
" \"two_omics_benchmark\": true,\n",
" \"use_conditionals\": false,\n",
" \"verbose\": 0,\n",
" \"view_dropout\": 0.3,\n",
" \"view_latent_dim\": 0.25,\n",
" \"view_loss_recon_type\": [\n",
" \"mean\",\n",
" \"mean\"\n",
" ],\n",
" \"view_loss_weights\": [\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0,\n",
" 1.0\n",
" ],\n",
" \"w_cat\": 0.01,\n",
" \"w_contrastive\": 0.005,\n",
" \"w_decay\": 0.0005,\n",
" \"w_gauss\": 0.0001,\n",
" \"w_kl\": 0.0001,\n",
" \"w_rec\": 1\n",
"}\n"
]
}
],
"source": [
"vae_imputed, vae_latent = train.load_vae_reconstructions()\n",
"vae_predicted, _ = train.load_vae_reconstructions(mode=\"all\")\n",
"\n",
"mofa_imputed, mofa_latent = CLinesDatasetMOFA.load_reconstructions(clines_db)\n",
"move_diabetes_imputed, move_diabetes_latent = (\n",
" CLinesDatasetMOVE.load_reconstructions(clines_db)\n",
")\n",
"jamie_imputed, jamie_latent = CLinesDatasetJAMIE.load_reconstructions(clines_db)\n",
"scvaeit_imputed, scvaeit_latent = CLinesDatasetSCVAEIT.load_reconstructions(clines_db)\n",
"\n",
"_, mixOmics_latent = CLinesDatasetMixOmics.load_reconstructions(clines_db)\n",
"_, iClusterPlus_latent = CLinesDatasetIClusterPlus.load_reconstructions(clines_db)\n",
"_, moCluster_latent = CLinesDatasetMoCluster.load_reconstructions(clines_db)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [],
"source": [
"samples_mgexp = ~clines_db.dfs[\"transcriptomics\"].isnull().all(axis=1)\n",
"\n",
"gexp_gdsc = pd.read_csv(f\"./data/clines/transcriptomics.csv\", index_col=0).T\n",
"gexp_mosa = vae_imputed[\"transcriptomics\"]\n",
"gexp_mosa_7omics = mosa_7omics_imputed[\"transcriptomics\"]\n",
"gexp_move = move_diabetes_imputed[\"transcriptomics\"]\n",
"gexp_jamie = jamie_imputed[\"transcriptomics\"]\n",
"gexp_mofa = mofa_imputed[\"transcriptomics\"]\n",
"gexp_scvaeit = scvaeit_imputed[\"transcriptomics\"]\n",
"\n",
"gexp_dfs = dict(\n",
" [\n",
" (\"MOSA_7omics\", gexp_mosa_7omics),\n",
" (\"MOSA_2omics\", gexp_mosa),\n",
" (\"MOFA\", gexp_mofa),\n",
" (\"MOVE\", gexp_move),\n",
" (\"JAMIE\", gexp_jamie),\n",
" (\"scVAEIT\", gexp_scvaeit),\n",
" ]\n",
")\n",
"samples = set(gexp_gdsc.index).intersection(gexp_mosa.index)\n",
"genes = list(set(gexp_gdsc.columns).intersection(gexp_mosa.columns))"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [],
"source": [
"gexp_corr_dfs = []\n",
"for name in gexp_dfs:\n",
" gexp_corr = pd.DataFrame(\n",
" [\n",
" two_vars_correlation(\n",
" gexp_gdsc.loc[s, genes],\n",
" gexp_dfs[name].loc[s, genes],\n",
" method=\"pearson\",\n",
" extra_fields=dict(sample=s, with_gexp=samples_mgexp.loc[s]),\n",
" )\n",
" for s in samples\n",
" ]\n",
" )\n",
" gexp_corr['model'] = name\n",
" gexp_corr_dfs.append(gexp_corr)\n",
"gexp_corr_dfs = pd.concat(gexp_corr_dfs)"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {},
"outputs": [],
"source": [
"_, ax = plt.subplots(1, 1, figsize=(1.5, 2), dpi=600)\n",
"\n",
"sns.boxplot(\n",
" data=gexp_corr_dfs[~gexp_corr_dfs[\"with_gexp\"]],\n",
" x=\"model\",\n",
" y=\"corr\",\n",
" hue=\"model\",\n",
" palette=\"tab10\",\n",
" linewidth=0.3,\n",
" fliersize=1,\n",
" notch=True,\n",
" saturation=1.0,\n",
" showcaps=False,\n",
" boxprops=dict(linewidth=0.5, edgecolor=\"black\"),\n",
" whiskerprops=dict(linewidth=0.5, color=\"black\"),\n",
" flierprops=dict(\n",
" marker=\"o\",\n",
" markerfacecolor=\"black\",\n",
" markersize=1.0,\n",
" linestyle=\"none\",\n",
" markeredgecolor=\"none\",\n",
" alpha=0.6,\n",
" ),\n",
" medianprops=dict(linestyle=\"-\", linewidth=0.5),\n",
" ax=ax,\n",
")\n",
"ax.set_xticklabels(ax.get_xticklabels(), rotation=45, ha=\"right\")\n",
"ax.set(\n",
" title=f\"\",\n",
" ylabel=\"Correlation between reconstructed\\nand GDSC transcriptomics (Pearson's r)\",\n",
" xlabel=f\"Sample without transcriptomics\\nduring training\",\n",
")\n",
"\n",
"PhenPred.save_figure(\n",
" f\"{plot_folder}/{hyperparameters['load_run']}_reconstructed_gexp_correlation_boxplot_2vs7omics\"\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [],
"source": [
"gexp_corr_out_sample_dfs = gexp_corr_dfs[~gexp_corr_dfs[\"with_gexp\"]]"
]
},
{
"cell_type": "code",
"execution_count": 50,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"ShapiroResult(statistic=0.9753528740783312, pvalue=0.00012429713994164323)"
]
},
"execution_count": 50,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"shapiro(\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_7omics'\")[\"corr\"],\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 53,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"TtestResult(statistic=21.43230547868578, pvalue=3.3627101251530766e-74, df=540.0)\n",
"WilcoxonResult(statistic=65.0, pvalue=6.947921853872344e-46)\n"
]
}
],
"source": [
"print(ttest_ind(\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_7omics'\")[\"corr\"],\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_2omics'\")[\"corr\"],\n",
"))\n",
"print(wilcoxon(\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_7omics'\")[\"corr\"],\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_2omics'\")[\"corr\"],\n",
"))"
]
},
{
"cell_type": "code",
"execution_count": 54,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"TtestResult(statistic=18.24570154229317, pvalue=2.6836218299727367e-58, df=540.0)\n",
"WilcoxonResult(statistic=74.0, pvalue=7.675247614114657e-46)\n"
]
}
],
"source": [
"print(\n",
" ttest_ind(\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_7omics'\")[\"corr\"],\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOFA'\")[\"corr\"],\n",
" )\n",
")\n",
"print(\n",
" wilcoxon(\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_7omics'\")[\"corr\"],\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOFA'\")[\"corr\"],\n",
" )\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 55,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"TtestResult(statistic=-2.282000726313907, pvalue=0.022877465018056837, df=540.0)\n",
"WilcoxonResult(statistic=7739.0, pvalue=1.2630246307937823e-16)\n"
]
}
],
"source": [
"print(\n",
" ttest_ind(\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_2omics'\")[\"corr\"],\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOFA'\")[\"corr\"],\n",
" )\n",
")\n",
"print(\n",
" wilcoxon(\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_2omics'\")[\"corr\"],\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOFA'\")[\"corr\"],\n",
" )\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 32,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"TtestResult(statistic=29.191437824112302, pvalue=1.6706086209243048e-85, df=270)"
]
},
"execution_count": 32,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"ttest_rel(\n",
" gexp_corr_out_sample_dfs.query(\"model == 'MOSA_7omics'\")[\"corr\"],\n",
" gexp_corr_out_sample_dfs.query(\"model == 'JAMIE'\")[\"corr\"],\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Drug"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [],
"source": [
"plot_df = pd.read_csv(\n",
" \"./reports/vae/drugresponse/20240830_110319_predicted_ctd2_corr.csv\"\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [],
"source": [
"plot_out_df = plot_df[plot_df[\"MOSA_outofsample\"] == \"Out-of-sample\"]\n",
"plot_in_df = plot_df[plot_df[\"MOSA_outofsample\"] == \"In-sample\"]"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"TtestResult(statistic=3.44106965740751, pvalue=0.0006309014889326985, df=472.0)\n",
"WilcoxonResult(statistic=3480.0, pvalue=8.939902145451377e-24)\n"
]
}
],
"source": [
"print(ttest_ind(\n",
" plot_out_df[plot_out_df[\"method\"] == \"MOSA_corr\"][\"corr\"],\n",
" plot_out_df[plot_out_df[\"method\"] == \"MOFA_corr\"][\"corr\"],\n",
"))\n",
"print(wilcoxon(\n",
" plot_out_df[plot_out_df[\"method\"] == \"MOSA_corr\"][\"corr\"],\n",
" plot_out_df[plot_out_df[\"method\"] == \"MOFA_corr\"][\"corr\"],\n",
"))"
]
},
{
"cell_type": "code",
"execution_count": 57,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"TtestResult(statistic=-0.10472369634495242, pvalue=0.916639544284535, df=472.0)\n",
"WilcoxonResult(statistic=13553.0, pvalue=0.6036728652783327)\n"
]
}
],
"source": [
"print(\n",
" ttest_ind(\n",
" plot_out_df[plot_out_df[\"method\"] == \"MOSA_corr\"][\"corr\"],\n",
" plot_out_df[plot_out_df[\"method\"] == \"MOSA_7omics_corr\"][\"corr\"],\n",
" )\n",
")\n",
"print(\n",
" wilcoxon(\n",
" plot_out_df[plot_out_df[\"method\"] == \"MOSA_corr\"][\"corr\"],\n",
" plot_out_df[plot_out_df[\"method\"] == \"MOSA_7omics_corr\"][\"corr\"],\n",
" )\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "mosa",
"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": "3.10.14"
},
"orig_nbformat": 4
},
"nbformat": 4,
"nbformat_minor": 2
}
Datasets
Models
