Protein language models learn the underlying accessibility of mutations alongside protein fitness landscapes
Oscar A. MacLean, Kieran lamb, Spyros Lytras, Laura Noelia Mojsiejczuk, Ke Yuan, Joseph Hughes, David L. Robertson
This study evaluates how protein language models capture mutational processes and fitness landscapes, finding that explicit mutational accessibility improves predictions while historical biases degrade performance.
bioRxiv · Jul 9, 2026 · PreprintEvolutionary Biology