Artificial intelligence guided discovery of a barrier-protective therapy in inflammatory bowel disease

Debashis Sahoo,Soumita Das,Pradipta Ghosh,Gajanan D Katkar,Stella-Rita Ibeawuchi,Ibrahim M Sayed,Lee Swanson,John T Chang,Rama F Pranadinata,Courtney Tindle,Dominik L Stec,Yash Mittal,Mackenzie Fuller,William J Sandborn

doi:10.1038/s41467-021-24470-5

Abstract

Modeling human diseases as networks simplify complex multi-cellular processes, helps understand patterns in noisy data that humans cannot find, and thereby improves precision in prediction. Using Inflammatory Bowel Disease (IBD) as an example, here we outline an unbiased AI-assisted approach for target identification and validation. A network was built in which clusters of genes are connected by directed edges that highlight asymmetric Boolean relationships. Using machine-learning, a path of continuum states was pinpointed, which most effectively predicted disease outcome. This path was enriched in gene-clusters that maintain the integrity of the gut epithelial barrier. We exploit this insight to prioritize one target, choose appropriate pre-clinical murine models for target validation and design patient-derived organoid models. Potential for treatment efficacy is confirmed in patient-derived organoids using multivariate analyses. This AI-assisted approach identifies a first-in-class gut barrier-protective agent in IBD and predicted Phase-III success of candidate agents.

Highlights

Modeling human diseases as networks simplify complex multi-cellular processes, helps understand patterns in noisy data that humans cannot find, and thereby improves precision in prediction
Boolean Network Explorer (BoNE)-enabled exploration of the Boolean paths (Supplementary Method; Fig. 2B; Supplementary Fig. S1D–F) revealed how some of the biggest clusters are connected by a series of BIRs (Green-Red arrows/Black-Blue lines, Fig. 2C)
A time series of Inflammatory Bowel Disease (IBD)-associated invariant events emerged— epithelial tight junctions (TJs) and other types of cellcell junctions appeared leftmost on the healthy side (C#1–2) of the IBD-map (Supplementary Data 2), levels of which are downregulated early during disease initiation and are progressively lost

Summary

Introduction

Modeling human diseases as networks simplify complex multi-cellular processes, helps understand patterns in noisy data that humans cannot find, and thereby improves precision in prediction. Target transcript analysis by quantitative PCR (qPCR) from human colon biopsies showed a significant decrease in PRKAB1 and a concomitant increase in CLDN2 expression in IBD-afflicted tissues, representing both UC and CD, regardless of disease location (Fig. 3E).

Results

Conclusion