A unified multimodal model for generalizable zero-shot and supervised protein function prediction

Predicting protein function is a fundamental yet challenging task that requires integrating diverse biological data modalities to capture complex functional relationships. Traditional machine learning methods often rely on single modalities or combine only a limited number (typically two), without aligning them in a unified representation, thereby constraining predictive accuracy. Moreover, most existing approaches are limited to preselected subsets of Gene Ontology (GO) function terms with sufficient annotations, making the prediction of novel function terms a persistent challenge. Here, we present FunBind, a multimodal AI model that jointly learns from protein sequences, textual descriptions, domain annotations, structural features, and GO terms to enhance prediction accuracy and infer previously unseen functions. FunBind operates in two modes: (1) self-supervised pretraining using contrastive learning to align the sequence modality with other heterogeneous modalities in a unified latent space, enabling unsupervised zero-shot function prediction, and (2) supervised fine-tuning that leverages all modalities for comprehensive and accurate function classification. Our results show that FunBinds zero-shot capabilities allow it to generalize effectively to novel function terms never encountered before, while its joint fine-tuning strategy substantially outperforms single-modality models and current state-of-the-art approaches in prediction accuracy.