Decoding sequence determinants of gene expression in diverse cellular and disease states

2026-05-25 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Artificial Intelligence & Machine Learning, Life Sciences & Biology, Health & Medical Research · Depth: Expert, long

Summary

Decima is a new sequence-to-function model that predicts cell type- and condition-specific gene expression directly from genomic DNA sequence. This model addresses limitations of prior approaches by training on single-cell or single-nucleus RNA sequencing data from over 22 million cells, enabling predictions at a much higher resolution. Decima successfully predicts the cell-type-specific expression of previously unseen genes. The model reveals *cis*-regulatory mechanisms that drive cell-type-specific gene expression and its changes in disease, predicts noncoding-variant effects at cell type resolution, and facilitates the design of context-specific regulatory DNA elements. Its code and model weights are publicly available via GitHub and Zenodo.

Key takeaway

For research scientists investigating gene regulation or disease mechanisms, Decima offers a powerful tool to move beyond bulk analysis. You can use this model to precisely identify *cis*-regulatory elements and predict noncoding variant effects within specific cell types and disease states. Consider integrating Decima into your workflow to design targeted regulatory DNA elements for gene therapy or functional genomics studies, enhancing the specificity of your interventions.

Key insights

Decima predicts cell-type-specific gene expression from DNA sequence, enabling precise regulatory mechanism and variant effect analysis.

Principles

• Single-cell data improves gene expression prediction resolution.
• *Cis*-regulatory elements drive cell-type identity and disease changes.
• Noncoding variants exhibit cell-type-specific regulatory effects.

Method

Decima is trained on >22 million single-cell/nucleus RNA-seq datasets to predict gene expression from surrounding DNA sequence, then used for attribution analysis and directed evolution.

In practice

• Analyze *cis*-regulatory mechanisms in specific cell types.
• Predict noncoding variant impacts at cellular resolution.
• Design context-specific regulatory DNA elements.

Topics

• Gene Expression Prediction
• Single-cell RNA Sequencing
• Cis-regulatory Elements
• Noncoding Variants
• Disease Genomics
• Regulatory DNA Design

Code references

• Genentech/decima
• Genentech/decima-applications

Best for: AI Scientist, Research Scientist

Related on AIssential

• Teaching an old dog new cells — Bulk-trained sequence models can be efficiently adapted for single-cell gene regulation and variant effect prediction.
• Network ontology transcript annotation identifies genetic signals underlying sex determination — Silver ion treatment in cannabis perturbs phytohormone signaling, inducing sex reversal by altering cell wall and embryogenic tissue genes.
• GenoMAS: A Multi-Agent Framework for Scientific Discovery via Code-Driven Gene Expression Analysis — GenoMAS automates gene expression analysis using a multi-agent LLM framework that balances structured workflows with adaptive, code-driven problem-solving.
• 🔬ESM: The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub — Scaling protein language models with vast, diverse data reveals emergent biological capabilities and enables programmable design.
• [R] Genomic Large Language Models — Genomic LLMs can identify functional similarities in DNA beyond sequence alignment, indicating learned regulatory patterns.
• Probabilistic modelling of single-cell bisulfite sequencing data with MethylVI — MethylVI disentangles biological signals from technical noise in single-cell bisulfite sequencing data.

Open in AIssential →

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