parsaidp/bioreason-proteinllm

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 15, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

BioReason ProteinLLM by parsaidp is a 4 billion parameter multimodal model based on Qwen3-4B, specifically fine-tuned for protein reasoning. It integrates a Qwen3-4B text model with an ESM3 protein encoder and a GAT-based Gene Ontology graph encoder. This model excels at protein function prediction by combining textual, sequence, and ontological information, offering a specialized tool for bioinformatics research.

Loading preview...

BioReason ProteinLLM Overview

BioReason ProteinLLM is a specialized 4 billion parameter multimodal model developed by parsaidp, designed for advanced protein function prediction. It uniquely combines a fine-tuned Qwen3-4B text model with a 1.4 billion parameter ESM3 small open v1 protein encoder and a GAT-based Gene Ontology (GO) graph encoder.

Key Capabilities

  • Multimodal Integration: Seamlessly processes and integrates information from protein sequences, textual descriptions, and Gene Ontology graphs.
  • Protein Function Prediction: Optimized for predicting protein functions by leveraging diverse biological data sources.
  • Specialized Architecture: Utilizes a Qwen3-4B base for reasoning, an ESM3 encoder for protein sequences, and a GAT for GO graph embeddings, with dedicated projection layers to bridge these modalities.

Good For

  • Researchers and developers in bioinformatics and computational biology.
  • Tasks requiring protein function inference based on sequence, text, and ontological context.
  • Applications benefiting from a multimodal understanding of proteins.