Combining explainable machine learning, demographic and multi-omic data to identify precision medicine strategies for inflammatory bowel disease

Abstract

AbstractInflammatory bowel diseases (IBDs), including ulcerative colitis and Crohn’s disease, affect several million individuals worldwide. These diseases are heterogeneous at the clinical, immunological and genetic levels and result from a complex interaction between the host and environmental factors. Investigating drug efficacy in cultured human fresh IBD tissues can improve our understanding of the reasons why certain medications are more or less effective for different patients.We propose an explainable machine learning (ML) approach that combines bioinformatics and domain insight, to informatively integrate multi-modal data to predict inter-patient specific variation in drug response. Using explanation of our models, we interpret the models’ predictions inferring unique combinations of important features associated with human tissue pharmacological responses. The inferred multi-modal features originate from multi-omic data (genomic and transcriptomic), demographic, medicinal and pharmacological data and all are associated with drug efficacy generated by a preclinical human fresh IBD tissue assay.To pharmacologically assess patient variation in response to IBD treatment, we used the reduction in the release of the inflammatory cytokine TNFα from the fresh IBD tissues in the presence or absence of test drugs, as a measure of drug efficacy. The TNF pathway is a common target in current therapies for IBD; we initially explored the effects of a mitogen-activated protein kinase (MAPK) inhibitor on the production of TNFα; however, we later show the approach can be applied to other targets, test drugs or mechanisms of interest. Our best model was able to predict TNFα levels from a combination of integrated demographic, medicinal and genomic features with an error as low as 4.98% on unseen patients. We incorporated transcriptomic data to validate and expand insights from genomic features. Our results showed variations in drug effectiveness between patients that differed in gender, age or condition and linked new genetic polymorphisms in our cohort of IBD patients to variation in response to the anti-inflammatory treatment BIRB796 (Doramapimod).Our approach models drug response in a relevant human tissue model of IBD while also identifying its most predictive features as part of a transparent ML-based precision medicine strategy.