Summary
Zero-shot variant effect prediction using sequence-only protein language models can be improved by averaging predictions from many orthologs (1). This was shown with ESM and ProGen.
Figures
Ref (1)
See also
- Inverse folding models can predict antibody-antigen binding affinities with higher accuracy by correcting with predictions from unbound state
- Zero-shot protein stability prediction using inverse folding models can be improved by subtracting predictions from residue in isolation
- Averaging logits from multiple sources can improve fitness prediction
1.
Pugh CWJ, Nuñez-Valencia PG, Dias M, Frazer J. From Likelihood to Fitness: Improving Variant Effect Prediction in Protein and Genome Language Models. openRxiv; 2025. Available from: https://doi.org/10.1101/2025.05.20.655154