Enhancing patient stratification and interpretability through class-contrastive and feature attribution techniques: unveiling potential therapeutic gene targets

Introduction

Code for startifying patients using class-contrastive machine learning on genomic data.

Files

Main files created from scratch:

• 1_gene_sampling_heatmaps.ipynb - sampling from the RISK dataset
• 2_preprocessing_training_autoencoder.ipynb - training the autoencoder
• 3_kmeans_gmm_eval.ipynb - clustering and classification using GMMs and KMeans, including evaluation
• gene_shap_minimal/4_gene_shap_dependent.ipynb - producing SHAP explainability plots to identify and rank risk genes by phenotype, using SHAP adapted for feature dependence
• gene_shap_minimal/4_gene_shap_independent.ipynb - producing SHAP explainability plots to identify and rank risk genes by phenotype, using original SHAP with feature independence
• gene_shap_minimal/5_consensus_clustering_gene_modules_deep_ulcer.ipynb - identification and characterisation of gene modules for Crohn's disease with deep ulcer
• gene_shap_minimal/5_consensus_clustering_gene_modules_no_ulcer.ipynb - identification and characterisation of gene modules for Crohn's disease without deep ulcer
• gene_shap_minimal/6_class_contrastive_GMM_gene_modules.ipynb - class-contrastive technique for cluster explainability based on identified gene modules
• gene_shap_minimal/6_class_contrastive_GMM_volcano_plot.ipynb - class-contrastive technique for cluster explainability based on gene sets extracted from volcano plot

This code builds upon the SHAP functionality. The original SHAP implementation can be found at https://github.com/slundberg/shap. A minimal version is provided using the .py files in gene_shap_minimal.
This file was modified to incorporate feature dependence into the method:

• gene_shap_minimal/explainers/_kernel.py

The following files were created from scratch for facilitation:

• gene_shap_minimal/kmeans_gmm_eval_fns.py - includes functions to get GMM predictions
• gene_shap_minimal/objects.py - functions to return required objects
• gene_shap_minimal/calc_shap.py - calculates and stores SHAP values

Installation

pip install -r requirements.txt

Manuscript

Forthcoming

Contact

Sharday Olowu and Soumya Banerjee

[email protected] [email protected]