ProteinGym

anchor

Findings

• IC-268ESM-2 and ProGen-2 zero-shot fitness prediction follows an inverted U-shape as a function of wild type sequence likelihood, with both under- and over-preferred sequences degrading performance [eval]
• IC-269Influence functions on ESM-2 650M reveal a power law tail in training data influence on sequence likelihood, with influence diminishing as Hamming distance from the wild type increases [eval]
• IC-270Unsupervised finetuning (evo-tuning) on homologous sequences improves ESM-2 650M zero-shot fitness prediction for low-likelihood wild types but harms high-likelihood ones, with optimal threshold at log-likelihood ε = −1.4 [eval]