EvoIF combines homolog profiles and inverse-folding signals for protein fitness prediction
KevinKaichuang · x · 2026-07-24
A repost challenges the idea that evolutionary data alone can predict a protein’s overall fitness, arguing that these models mostly learn whether a sequence looks viable in nature, not whether it is good for a specific target property.
The quoted paper summary introduces EvoIF, a lightweight predictor that combines two signals: within-family evolutionary profiles from homologs and a cross-family structural signal distilled from inverse-folding logits. The authors also offer an interpretation for why masked language modeling helps zero-shot fitness prediction: evolution can be seen as implicit reward maximization, with extant sequences acting as expert demonstrations.
Related event: EvoIF Model Enhances Protein Fitness Prediction(2 posts)→
More from Research
- Meta FAIR's Self-Direction turns agent failures into reusable LoRA skills, +13.8pp on Terminal-Bench — karen_ullrich · 2026-10-27
- DONIC hits 86.79% on bimanual dexterous tracking, crushing prior SOTA's 9.43% — liuziwei7 · 2026-10-27
- Why don't more LLMs ship 10M-token context? A researcher asks about attention tradeoffs — agarwl_ · 2026-10-27
- PSNR below 15 dB in papers likely means the metric was misused, researcher warns — kwangmoo_yi · 2026-10-27
- Ortet's first health eval: 34 open-weight models, 50 tasks, generic rankings fall short — thegautamkamath · 2026-10-27
- Scaling-fit discrepancy: output-head FLOPs are ~65% of compute at the smallest rung, ~3% at 535B — dlwh · 2026-10-27