Protein language model CLSS unifies sequence and structure in shared map

An international research team including the Earth-Life Science Institute at the Institute of Science Tokyo has published a protein language model that places a protein's amino acid sequence and its three-dimensional structure at the same location on a shared map. The model, Contrastive Learning Sequence-Structure, or CLSS, uses contrastive learning to produce similar embeddings for sequence-structure pairs while separating unrelated pairs. The researchers report CLSS reproduced relationships recorded in the expert-curated ECOD and CATH classification systems even though those classifications were not provided during training, and that short sequence fragments can often be positioned meaningfully alongside complete sequences and structures. The findings appear in Proceedings of the National Academy of Sciences.