Response-aware molecular subtyping of ulcerative colitis for patient stratification via interpretable machine learning.

What To Know

This study describes an interpretable machine-learning framework that creates “response-aware” molecular subtypes of ulcerative colitis by training on cohorts with known infliximab response, then applying the learned subtype model to other datasets.

Four subtypes were reported with different immune and molecular profiles and varying infliximab response rates; one subtype showed relatively low immune activation and the highest response rate.

The authors used SHAP values from a treatment-response predictive model to build representations for spectral clustering, identified subtype-associated biomarkers, and validated subtype patterns across independent cohorts. The approach is positioned as a way to stratify patients by likely biologic response and to generate hypothesis-driving biomarkers.

The paper is framed as a methodological advance (interpretable ML + response-aware subtyping) rather than a clinical trial showing treatment benefit. It suggests potential utility for patient stratification and biomarker development but does not itself change treatment guidelines or provide validated clinical decision rules.