Bio ML critique: many published models score within ~10% of a naive linear baseline

iskander · x · 2026-09-08

iskander (George Ho) observes that many published bio ML models score within roughly 10% of a naive linear baseline.

The short note points to a familiar pitfall in bio ML papers: models are benchmarked against deep learning baselines rather than simple linear regressions, making seemingly sophisticated models look better than they are.

Original post →

More from Research

Research channel →