Diffusion model generating regulatory DNAs

2026-02-11 · Source: Machine learning : nature.com subject feeds · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Life Sciences & Biology · Depth: Advanced, quick

Summary

A new generative AI model named DNA-Diffusion, developed by Luca Pinello's team at Harvard Medical School, can generate 200-bp DNA sequences with regulatory potential. Published on February 11, 2026, this diffusion model was trained on extensive DNase I hypersensitive site index data from three distinct cell lines. It learns to reconstruct the noise addition process to generate functional DNA sequences, conditional on a specified cell type. The generation process is controlled by a classifier-free guidance parameter, which balances sequence diversity and activity. In silico evaluations demonstrate that DNA-Diffusion generates sequences with high regulatory activities, including gene expression activation, without merely memorizing training data. Further analysis of these sequences reveals enriched regulatory motifs, such as transcription factor binding sites, offering mechanistic insights.

Key takeaway

For AI Researchers and Synthetic Biologists exploring novel genetic constructs, DNA-Diffusion offers a method to generate functional regulatory DNA sequences. Your team can leverage this approach to design targeted genetic elements for specific cell types, potentially accelerating gene therapy development or synthetic biology applications. Consider integrating diffusion models into your workflow for sequence design to explore diverse and active regulatory elements.

Key insights

DNA-Diffusion is a diffusion model generating functional regulatory DNA sequences conditional on cell type.

Principles

• Diffusion models can generate functional biological sequences.
• Classifier-free guidance balances diversity and activity in generation.

Method

DNA-Diffusion learns to reconstruct noise addition to generate 200-bp regulatory DNA sequences, conditioned on cell type, after training on DNase I hypersensitive site index data.

In practice

• Generate synthetic regulatory DNA for specific cell types.
• Identify novel regulatory motifs in generated sequences.

Topics

• Diffusion Models
• Synthetic Biology
• Gene Regulation
• Genomic Sequences

Best for: AI Researcher, AI Scientist, Research Scientist

Related on AIssential

• AI can write genomes — how long until it creates synthetic life? — The Evo2 genomic language model, an AI designed to read, interpret, and generate DNA, RNA, and protein sequences, marks a significant advancement in synthetic biology by designing whole genome sequenc…
• Sampling sea state using a diffusion model — Diffusion models offer a computationally efficient path to probabilistic global sea state estimation.
• Biological Reasoning-Informed Regression for Interpretable Regulatory DNA Activity Prediction — Integrating structured biological knowledge with LLMs via two-stage training improves interpretable DNA regulatory activity prediction.
• Leap in DNA Synthesis Slashes Time to Build New Genetic Sequences — Sidewinder enables rapid, high-fidelity DNA synthesis, resolving a key bottleneck for generative AI in biology.
• [R] Genomic Large Language Models — Genomic LLMs can identify functional similarities in DNA beyond sequence alignment, indicating learned regulatory patterns.
• PromoterAtlas: decoding regulatory sequences across Gammaproteobacteria using a transformer model — PromoterAtlas is a transformer model that decodes bacterial regulatory sequences across thousands of Gammaproteobacteria species.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.