ProTrek: Navigating the Protein Universe through Tri-Modal Contrastive Learning

Abstract ProTrek, a tri-modal protein language model, enables contrastive learning of protein s equence, s tructure, and f unction (SSF). Through its natural language search interface, users can navigate the vast protein universe in seconds, accessing nine distinct search tasks that cover all possible pairwise combinations of SSF. Additionally, ProTrek serves as a general-purpose protein representation model, excelling in various downstream prediction tasks through supervised transfer learning, thereby providing extensive support for protein research and analysis.