Discover how Google DeepMind’s AlphaGenome 2 uses AI and 107 million cell profiles to predict DNA effects, identify genetic variants, and advance disease research.
AlphaGenome 2 could help researchers understand why the same genetic variant affects different cells in different ways.
Google DeepMind’s latest DNA prediction model combines single-cell biology with AI to identify disease-related genetic changes, offering pharmaceutical researchers a more precise way to investigate gene regulation and potential therapeutic targets.
Image
What Is Cell-Specific DNA Analysis and Single-Cell Biology?
Almost every cell in the human body carries the same DNA. Yet a heart muscle cell, a neuron, and an immune cell behave very differently. Each activates different genes to perform its particular function.
This is what matters when researchers investigate genetic disease. A DNA variant may have little effect across most tissues but cause significant disruption in one specialised cell population.
Traditional genetic analysis can struggle to identify these differences. Genome-wide association studies (GWAS) often identify regions associated with disease, but they do not always reveal the exact genetic change responsible, the affected gene, or the cell type where the disruption occurs.
Understanding where a genetic variant acts can be as important as identifying the variant itself.
This is the problem AlphaGenome 2 wants to solve.
Announced by Google DeepMind on 11 November 2026, the model builds on the original AlphaGenome system by predicting the effects of DNA variations within specific cell types and biological states.
The model draws on more than 107 million individual cell profiles across 68 tissues, combining human and mouse biological data to improve its understanding of gene regulation.
How Does AlphaGenome 2 Predict DNA Effects?
The central development with AlphaGenome 2 is a change in how the model processes biological information.
The original AlphaGenome analysed long DNA sequences and generated predictions across multiple tissues and experimental measurements.
Although this approach provided broad coverage, it could struggle to capture subtle differences between closely related cell types.
AlphaGenome 2 uses a conditional architecture. Researchers provide the DNA sequence alongside a specific biological condition, such as a tissue, cell type, or cellular state.
The AlphaGenome model then predicts how that particular biological context influences the effects of the DNA sequence.
That’s important because the same genetic variant can produce different biological outcomes depending on the cell in which it appears.
The model can also work with newly generated single-cell data without requiring researchers to retrain it from scratch.
Its cell encoder converts gene-activity measurements into a representation of the biological state, allowing the model to investigate previously unseen cell populations.
Image
What biological measurements can the AlphaGenome 2 model predict?
AlphaGenome 2 examines several molecular processes involved in gene regulation, including:
Gene expression: How actively cells use particular genes.
Chromatin accessibility: Whether DNA regions are accessible to the cellular machinery that controls gene activity.
Histone modifications: Chemical changes to proteins around which DNA is organised, influencing gene regulation.
DNA methylation: Chemical modifications that can affect whether genes are switched on or off.
Another useful capability is cross-modality imputation. If researchers provide one type of experimental measurement, the model can estimate other unmeasured properties.
For example, gene-expression data could help predict chromatin accessibility in the same cellular context. This may help research teams decide which laboratory experiments deserve priority.
“This is one of the riskiest industries there is because you're dealing with human lives, you're dealing with experimental protocols, and you're dealing with regulatory bodies where you might not get another shot at that clinical trial."
Image
What Do AlphaGenome 2’s Results Reveal?
The strongest performance findings for AlphaGenome 2 concern the model’s ability to identify causal genetic variants and predict their effects on gene expression.
Google DeepMind evaluated AlphaGenome 2 using the TenK10K dataset, which maps 28 primary immune cell types.
The company reported that the model identified causal variants in 30% of cases, compared with 16% for the original AlphaGenome fine-tuned using single-cell data and other specialised methods.
It also improved predictions of whether a genetic variant increased or decreased gene activity. The reported directional accuracy increased from 23% to 43%.
The findings suggest that cell-specific modelling can improve the identification of genetic changes that broader tissue-level predictions might overlook.
However, the figures need careful interpretation. The 30% result does not mean the model can identify 30% of all disease-causing mutations in clinical practice but it reflects performance within a defined research dataset and evaluation framework.
Connecting genetic variants to inflammatory bowel disease
One particularly relevant example concerns inflammatory bowel disease (IBD).
Previous research identified a genetic variant affecting a regulatory region that controls ETS2, a gene involved in immune-cell activity. The variant increases ETS2 activity in gut macrophages, a specialised immune-cell population.
DeepMind reports that AlphaGenome 2 reproduced this finding computationally. It identified the relevant genetic change and predicted its selective effect on ETS2 in gut macrophages.
This demonstrates how the model could help researchers move from a broad genetic association towards a more specific biological explanation.
For drug discovery teams, that could help prioritise disease mechanisms, identify relevant cell populations, and investigate potential therapeutic targets.
Image
What Could AlphaGenome 2 Mean for Pharmaceutical R&D?
AlphaGenome 2 also builds on a wider development in DeepMind’s genomic AI research.
As Pharmatica explored in our earlier Insights into the AlphaGenome Atlas, the platform that maps the predicted regulatory effects of all nine billion possible single-nucleotide variants across the human genome, giving researchers a way to prioritise potentially disease-relevant genetic changes at unprecedented scale.
While the AlphaGenome Atlas provides a searchable resource for exploring variant impact, AlphaGenome 2 advances the underlying predictive approach by incorporating cell-specific biological context.
Together, these developments point towards a more connected approach to genomic research, helping scientists move from identifying potentially important DNA variants to investigating how their effects differ across cell types and biological conditions.
Neither removes the need for experimental confirmation, but both could strengthen the evidence used to prioritise targets and guide early pharmaceutical research.
Specifically, AlphaGenome 2 has the ability to connect genetic variation with specific cell populations with several potential applications across pharmaceutical research.
For any research team, the potential value lies in improving the quality of early research decisions.
Better biological understanding could support target selection and help teams avoid pursuing mechanisms that lack convincing evidence.
1. Improving genetic target identification
Researchers could use AlphaGenome 2 to investigate whether disease-associated variants influence a particular gene in a specific cell population.
This could help narrow the search for disease-driving mechanisms before committing resources to target validation.
2. Supporting precision medicine research
Some diseases involve different biological mechanisms across patient populations and cell states. Better predictions of cell-specific gene regulation could help researchers investigate why individuals respond differently to disease processes or treatments.
3. Prioritising experimental research
Laboratory experiments require time, specialist expertise, and resources. By predicting additional molecular measurements from existing single-cell data, AlphaGenome 2 could help R&D teams select the experiments most likely to resolve important biological questions.
4. Connecting genetics with therapeutic development
The AlphaGenome 2 model could help bridge the gap between genetic association studies and actionable drug discovery hypotheses.
Researchers may be able to identify which genes, regulatory regions, and cell types deserve closer investigation.
These applications are particularly relevant to complex diseases, where several cell populations and regulatory mechanisms can contribute to disease progression.
AI in Pharma: Why the Future of Healthcare Starts With Patients, Not Tech
Kate O’Reilly, President & Chair of the Healthcare Businesswomen’s Association (HBA) Dublin-Ireland Chapter & Healthcare Transformation Partner at Roche, discusses AI in pharma, patient engagement, and the future of healthcare innovation.
Image
What Are the Potential Limitations of AlphaGenome 2?
Despite its reported performance improvements, Google DeepMind explicitly states that neither AlphaGenome nor AlphaGenome 2 has been validated or approved for clinical or diagnostic purposes.
The model’s predictions therefore cannot replace laboratory confirmation, functional studies, or clinical evidence.
Several considerations will shape AlphaGenome 2’s eventual practical value:
Benchmark performance: Results from TenK10K need to be assessed alongside independent evaluations using other datasets and biological conditions.
Data representation: Training data quality and coverage influence which cell types, genetic backgrounds, and biological states a model can represent reliably.
Experimental confirmation: Predicted variant effects still require appropriate biological experiments to establish their significance.
Commercial access: Non-commercial researchers can access the model through the AlphaGenome API, while commercial research access is currently in private preview.
These considerations matter for research teams assessing whether to incorporate the model into existing research workflows.
The technology also raises broader questions about AI governance, data provenance, reproducibility, and the interpretation of computational evidence.
As models become more capable of generating biological predictions, research teams will need clear processes for testing those predictions independently.
Pharmatica has previously explored these challenges for AI-driven drug discovery, including why predictive performance alone cannot establish scientific reliability.
From Genetic Prediction to Better Biological Understanding
AlphaGenome 2 represents an important development in the relationship between AI, single-cell biology, and human genetics.
The platform’s ability to examine DNA effects within specific cellular contexts could help researchers identify disease mechanisms that broader genomic analysis overlooks.
The improvement in causal variant identification is encouraging, particularly when combined with its ability to interpret previously unseen cell states and predict additional molecular measurements.
However, better prediction is not the same as demonstrated therapeutic benefit. Research teams will need independent evidence and experimental confirmation before using these findings to support consequential development decisions.
At Pharmatica, we examine the technologies, scientific evidence, and strategic decisions shaping Pharmaceutical R&D. Our Insights focus on understanding where AI can improve research decisions, where its limitations remain, and how emerging capabilities could influence the development of future medicines.
Explore more Insights into the technologies shaping life sciences by joining our community.
Pharmatica: Insight. Connection. Impact.
Frequently Asked Questions
What is AlphaGenome 2?
AlphaGenome 2 is an AI model developed by Google DeepMind that predicts how DNA sequences and genetic variants affect gene regulation in specific cell types and biological states.
How does AlphaGenome 2 improve genetic variant prediction?
It uses a conditional modelling architecture that incorporates the target cell type or biological state alongside DNA sequence information. This helps it identify cell-specific genetic effects more accurately.
How accurate is AlphaGenome 2 at identifying causal variants?
On the TenK10K dataset, AlphaGenome 2 identified 30% of causal variants, compared with 16% for the previous comparison methods reported by Google DeepMind.
Can AlphaGenome 2 help pharmaceutical companies discover drug targets?
Yes. Its predictions could help researchers identify disease-associated genes, understand cell-specific biological mechanisms, and prioritise potential therapeutic targets for experimental investigation.
Is AlphaGenome 2 approved for clinical or diagnostic use?
No. AlphaGenome 2 is intended for research use only. Google DeepMind has not validated or approved it for clinical or diagnostic applications.
Nicole (BSc Molecular Medicine, Honours Medical Biochemistry) has many years of pharmaceutical experience, having worked for top CROs and biopharma companies for more than a decade.
Did you enjoy the content?
1
Why not support
Nicole Dale
by giving this content a like
Pharmacovigilance Trends Shaping Drug Safety Reporting in 2026
We explore the pharmacovigilance trends shaping drug safety in 2026, from AI and real-world data to ICH E2D(R1), to reporting quality for MedSafety Week 2026.
SynthID Bio embeds hidden watermarks in AI-designed proteins. See how Google DeepMind’s tool could strengthen biosecurity and what pharma R&D should do next.
The UK’s AI Biosecurity Lab Puts Pharma at the Centre of Security
The UK AI Biosecurity Lab brings government, AI developers, and life sciences together to detect biological threats and accelerate medical countermeasures.
White House Tightens Nucleic Acid Screening Policy
The U.S. nucleic acid synthesis screening policy sets new supplier, customer and equipment controls for pharma and biotech research governance and procurement.
Comments (0)