- Input: one or more protein sequences, optionally with masked positions to fill in.
- Output: sequence embeddings, per-position amino acid probabilities, sampled mutations, or naturalness scores.
Masked Models
Masked Models
Protein language models for embeddings, scoring, and sampling
Protein language models trained to fill in masked positions using surrounding context from
both directions. They produce sequence embeddings, per-position amino acid probabilities,
sampled mutations, and naturalness scores. These outputs are sequence-level priors, useful
for representation, local editing, and ranking rather than structural or functional
validation.
Oxford Protein Informatics Group (OPIG)
Meta AI
Biohub
