Google DeepMind has released AlphaGenome-2, a genomics AI model that predicts gene expression at single-cell type resolution. The model shifts from bulk tissue predictions to cell-type-specific analysis, enabling researchers to understand how genetic variants affect individual cell types.
The new model builds on AlphaGenome-1 and was evaluated across 127 tissue and cell type contexts. It represents a major step toward understanding the functional impact of genetic variation in human health and disease.
AlphaGenome-2 Moves Beyond Bulk Tissue Analysis
Traditional genomics models analyze bulk tissue samples, which average signals across many cell types. AlphaGenome-2 instead predicts gene expression levels for individual cell types within tissues.
This distinction matters because disease-causing variants often affect only specific cell types. A variant linked to heart disease, for example, may disrupt gene expression only in cardiomyocytes, not in the entire heart tissue sample.
The model’s single-cell resolution also helps identify which cell types are most vulnerable to specific genetic perturbations. This information is critical for developing targeted therapies.
Why Cell-Type Specificity Matters
Most disease-associated variants identified by genome-wide association studies (GWAS) are non-coding. They do not change protein sequences but instead affect gene regulation in subtle ways.
Understanding which cell types are affected by these variants is essential for deciphering disease mechanisms. AlphaGenome-2 provides this cell-type-level insight directly from DNA sequence.
How AlphaGenome-2 Works
AlphaGenome-2 uses a context-based approach to predict gene expression. The model takes a DNA sequence and a specific cell type context as input, then predicts how that sequence drives gene expression in that context.
The model was trained on data from the Genotype-Tissue Expression (GTEx) project and other genomic resources. It learns which regulatory elements are active in which cell types.
The key innovation is the ability to predict expression across 127 different contexts, from whole tissues down to specific cell types.
Predicting Variant Effects
The model can also predict the impact of genetic variants on gene expression. This enables researchers to prioritize variants that are most likely to be pathogenic.
By comparing predicted expression levels between reference and alternative alleles, AlphaGenome-2 can flag variants that disrupt gene regulation in specific cell types. This is particularly valuable for interpreting variants of uncertain significance in clinical genomics.
Why Single-Cell Resolution Matters
Single-cell resolution helps bridge the gap between genetic association and biological mechanism. GWAS identify variants linked to diseases, but they do not explain how those variants act.
AlphaGenome-2 helps close this gap by showing which cell types are affected by which variants. This can reveal the cell types most relevant to a disease and suggest therapeutic targets.
The model also helps interpret non-coding variants, which make up the majority of disease-associated genetic variation but are difficult to study experimentally.
Key Capabilities of AlphaGenome-2
- Cell-type-specific expression prediction: The model predicts gene expression levels for individual cell types, not just bulk tissues.
- Variant effect prediction: It can assess how genetic variants alter gene expression in specific cell types.
- Context-based learning: The model uses cell type context as an input, allowing flexible predictions across many contexts.
- Broad coverage: It was evaluated across 127 tissue and cell type contexts, covering a wide range of human biology.
Availability and Access
AlphaGenome-2 is available through Google Cloud, allowing researchers to run predictions at scale. DeepMind has positioned the model as a resource for the broader genomics community.
The model is part of DeepMind’s broader push into biology, following AlphaFold’s success in protein structure prediction. AlphaGenome-2 extends this work into the regulatory genome.
Limitations and Next Steps
The model’s predictions are computational and require experimental validation. Researchers will need to confirm AlphaGenome-2’s predictions in the lab before drawing clinical conclusions.
The current version focuses on gene expression prediction. Future versions may incorporate additional regulatory mechanisms, such as chromatin state and epigenetic modifications.
AlphaGenome-2 represents a shift from “what variants are associated with disease” to “which cell types and mechanisms are involved.”
This shift could accelerate drug discovery and precision medicine by identifying the right cell types to target for therapeutic intervention.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.