Summary

Attention matrices calculated from antibody PLMs can be used for paratope prediction (1). This was done using AntiBERTy, and authors attribute this to the use of multiple instance learning, which teaches the model to discriminate between binding and non-binding sequences via overrepresentation of binders in Immune repertoires.

See also

1.
Ruffolo JA, Gray JJ, Sulam J. Deciphering antibody affinity maturation with language models and weakly supervised learning. 2021; Available from: https://arxiv.org/abs/2112.07782