AI-Powered Analysis to Identify Potential Patient-Specific Therapy for Crohn's Disease Using Single-Cell RNA Sequencing

Abstract

Inflammatory bowel disease (IBD), affecting over 3 million Americans, describes chronic inflammatory disorders of the digestive tract, with Crohn's disease as a major subtype affecting the small intestine and colon. Patients often cycle through multiple therapies including corticosteroids, 5-aminosalicylates, immunomodulators, anti-TNF agents, anti-integrin antibodies, and JAK inhibitors before achieving remission. This trial-and-error approach prolongs disease activity, reduces quality of life, and burdens patients and the economy. The current treatment paradigm relies on population-level responses rather than patient-specific biology, therefore precision medicine approaches are needed.This project developed an AI-powered pipeline (automated sequential steps) identifying treatments by combining single-cell RNA sequencing, which measures gene activity in individual cells, with machine learning. Using data from 347,017 cells across multiple Crohn's disease patients, gene set enrichment analysis, which identifies abnormally active cellular processes, was performed across 2,186 pathways from KEGG, Hallmark, and Reactome databases. Singular value decomposition, which mathematically groups related pathways by similarity, compressed them into 20 summary patterns. Random Forest classifiers were trained for both tissues (small intestine and colon) to score treatment compatibility.The pipeline distinguished effective from ineffective treatment matches across 480 pathway-patient pairs. The small intestine model achieved 73.8% accuracy and area under the curve (AUC) of 0.811, while the colon model achieved 75.7% accuracy and 0.767, where 0.5 is random chance. All 480 recommendations received high confidence scores above 0.7.This pipeline offers a data-driven path toward treatments matched to each patient's unique biology, potentially transforming Crohn's disease treatment.