Biological Reasoning-Informed Regression for Interpretable Regulatory DNA Activity Prediction

2026-06-06 · Source: Machine Learning · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

R3LM is a novel framework designed to enhance the prediction of DNA cis-regulatory element (CRE) activity by integrating large language models (LLMs) with structured biological knowledge. Addressing the limitations of black-box regression methods and direct LLM application to raw DNA, R3LM introduces a biologically grounded data format to improve LLM understanding of regulatory information. It also presents CRE-ReasonBench, the first dataset linking DNA sequences and activity scores with mechanistic reasoning traces. Through a two-stage training process—first teaching LLMs reasoning over structured biological information, then performing regression—R3LM achieves state-of-the-art performance in enhancer prediction across three cell types. This framework surpasses both LLMs using raw sequence input and specialized DNA models, offering interpretable mechanistic explanations and serving as a potential interpretable reward model for biologists in CRE design.

Key takeaway

For biologists and AI scientists developing models for gene regulation or designing cis-regulatory elements, R3LM offers a path to highly accurate and interpretable predictions. You should consider adopting structured biological knowledge and two-stage LLM training to move beyond black-box models. This approach can significantly improve your ability to understand mechanistic explanations and refine CRE designs.

Key insights

Integrating structured biological knowledge with LLMs via two-stage training improves interpretable DNA regulatory activity prediction.

Principles

• Structured biological data enhances LLM understanding.
• Mechanistic reasoning improves prediction interpretability.
• Two-stage training optimizes LLM for specific tasks.

Method

R3LM uses a biologically grounded data format and CRE-ReasonBench for two-stage LLM training: first, reasoning over structured biological information, then performing regression for DNA regulatory activity prediction.

In practice

• Design interpretable reward models.
• Predict enhancer activity across cell types.
• Structure DNA data for LLM input.

Topics

• DNA Regulatory Elements
• Gene Expression Prediction
• Large Language Models
• Biological Knowledge Integration
• Interpretable AI
• Enhancer Prediction

Code references

• DuanYi516/R3LM

Best for: AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• [R] Genomic Large Language Models — Genomic LLMs can identify functional similarities in DNA beyond sequence alignment, indicating learned regulatory patterns.
• Multi-to-uni modal knowledge transfer pre-training for molecular representation learning — M2UMol enables robust molecular representation learning from incomplete multimodal data by transferring knowledge to a 2D encoder.
• Learning the PTM code through a coarse-to-fine mechanism-aware framework — COMPASS-PTM unifies PTM site prediction and enzyme assignment by modeling PTM patterns and catalytic regulators.
• Opening new paths in aging research — AI tools like Co-Scientist can synthesize complex biological literature to generate novel, testable hypotheses.
• Knowledge Graphs and Reasoning LLMs for Finding Simple Yet Effective Transcriptomic Perturbation Predictors — Simple K-nearest neighbour models leveraging knowledge graphs, enhanced by RL-trained LLMs, offer competitive transcriptomic perturbation prediction.
• Apr 29, 2026ScienceEvaluating Claude’s bioinformatics research capabilities with BioMysteryBench — Claude models are rapidly advancing in bioinformatics, matching or exceeding human expert performance on complex, real-world tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.