Summary
Protein language models have an implicit understanding of the uncertainty of their predictions. This can be measured through pseudolikelihood and perplexity structure in the model’s sequence predictions (1).
Figures
Ref (1)
See also
- Sequence perplexity
- Uncertainty quantification
- High-confidence predictions from protein language models co-cluster together in embedding space and correlate with performance on variant effect prediction tasks
1.
Bigot A, Bhasin H, Park CF, Shakhnovich E, Wang D. Viral Proteins Reveal Geometry of Protein Language Models. arXiv; 2026. Available from: https://arxiv.org/abs/2606.12609