Hebrew University of Jerusalem

Contrastively fine-tuned ESM-2 (35M and 150M) protein language models that produce general-purpose sequence embeddings where biological similarity maps to embedding proximity.

Dual-encoder contrastive model that embeds protein domains and peptides into a shared space to predict domain-peptide binding specificity at proteome scale.

Universal protein language model pretrained on 106M UniRef90 sequences with dual objectives: masked language modeling and Gene Ontology annotation prediction.