BioReason-Pro: Multimodal reasoning AI for protein function prediction
Built on GO-GPT, a model that predicts Gene Ontology terms with state-of-the-art accuracy, BioReason-Pro integrates protein embeddings from the ESM3 foundation model with the Qwen3 language model to generate step-by-step reasoning traces connecting domain architecture, interaction partners, and organism context to molecular function, biological processes, and cellular localization. The system is further optimized via reinforcement learning against proteins with known experimental annotations. Rather than returning a label, it explains its logic, so researchers can evaluate, interrogate, and decide whether a prediction warrants experimental follow-up.
Github