<ul><li>Prot2Token is a unified framework for protein modeling that converts various protein-related predictions into a standardized next-token prediction format.</li><li>It employs an autoregressive decoder conditioned on embeddings from pre-trained protein encoders and guided by learnable task tokens to perform diverse predictions.</li><li>The architecture allows for multi-task learning, enabling a single model to excel in multiple protein prediction tasks with improved efficiency.</li><li>Extensive experimental validation shows that Prot2Token offers strong predictive power, significant speedups, and performance matching or exceeding specialized approaches, which can accelerate biological discovery and therapeutic development.</li></ul>

Prot2Token: A Unified Framework for Protein Modeling via Next-Token Prediction

Discover more