High-confidence predictions from protein language models co-cluster together in embedding space and correlate with performance on variant effect prediction tasks

Summary

High-confidence predictions (assessed with pLDDT) from protein language models co-cluster together in embedding space and their placement therein correlates with performance on variant effect prediction tasks (1). In contrast, low-confidence predictions fall in a “junkyard”.

Figures

Figures from (1)

1.

Prabakaran R, Bromberg Y. Quantifying uncertainty in protein representations across models and tasks. Nature Methods. 2026;23(4):796–804. Available from: https://doi.org/10.1038/s41592-026-03028-7