BioReason: Incentivizing Multimodal Biological Reasoning within a DNA-LLM Model

2025-05-29

AI Summary

Existing DNA foundation models exhibit strong sequence representation capabilities but suffer from limited multi-step biological reasoning and lack transparent, interpretable internal mechanisms. To address these limitations, we propose DNA-LLM, a cross-modal collaborative architecture that enables large language models (LLMs) to perform end-to-end, multi-step, and interpretable reasoning directly over raw DNA sequences—the first of its kind. Our method integrates DNA foundation models with LLMs, incorporates biologically grounded reinforcement learning, and introduces step-level mechanistic attribution for traceable reasoning. Moreover, it supports zero-shot generalization to unseen biological entities. Evaluated on KEGG disease pathway prediction, DNA-LLM achieves 97% accuracy (+9% absolute improvement), and delivers an average 15% gain across tasks including variant effect prediction. All code, datasets, and model weights are publicly released.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Unlocking deep, interpretable biological reasoning from complex genomic data is a major AI challenge hindering scientific discovery. Current DNA foundation models, despite strong sequence representation, struggle with multi-step reasoning and lack inherent transparent, biologically intuitive explanations. We introduce BioReason, a pioneering architecture that, for the first time, deeply integrates a DNA foundation model with a Large Language Model (LLM). This novel connection enables the LLM to directly process and reason with genomic information as a fundamental input, fostering a new form of multimodal biological understanding. BioReason's sophisticated multi-step reasoning is developed through supervised fine-tuning and targeted reinforcement learning, guiding the system to generate logical, biologically coherent deductions. On biological reasoning benchmarks including KEGG-based disease pathway prediction - where accuracy improves from 88% to 97% - and variant effect prediction, BioReason demonstrates an average 15% performance gain over strong single-modality baselines. BioReason reasons over unseen biological entities and articulates decision-making through interpretable, step-by-step biological traces, offering a transformative approach for AI in biology that enables deeper mechanistic insights and accelerates testable hypothesis generation from genomic data. Data, code, and checkpoints are publicly available at https://github.com/bowang-lab/BioReason

Problem

• Enabling deep interpretable reasoning from genomic data
• Integrating DNA foundation models with LLMs for biological understanding
• Improving accuracy in disease pathway and variant prediction

Innovation

• Integrates DNA foundation model with LLM
• Uses supervised fine-tuning and reinforcement learning
• Improves accuracy in biological reasoning benchmarks