Show HN: Foundation models that predict patient response in clinical trials

Category: ai-ml

Tags: clinical-trials, foundation-model, bioinformatics

Score: 7.3/10 (Innovation: 7, Technical: 8, Documentation: 6, Utility: 7)

This project applies foundational models of patient biology to predict clinical trial responses from baseline biopsies, demonstrating improved sample efficiency and a concrete case study achieving AUROC 0.76 for ustekinumab in ulcerative colitis. It is interesting because it aims to reduce trial size and cost by leveraging pretrained embeddings across millions of samples, addressing a long-standing bottleneck in drug development.

Target audience: data scientists, drug developers, clinical researchers

Repository: https://atlasdiscovery.bio/clinical-trial-response-prediction

View on Hacker News