Google DeepMind announced AlphaGenome 2, an AI system trained on over 100 million cell profiles that predicts how genetic variants affect specific cell types, potentially advancing disease research and therapeutic design.
AlphaGenome 2 introduced with cell‑type specific predictions
On September 20, Google DeepMind released AlphaGenome 2, the successor to the AlphaGenome model launched last year. The new system is engineered to assess how DNA mutations influence each cell type and state, from cardiac muscle cells to immune cells. By modelling distinct gene‑expression patterns, AlphaGenome 2 delivers a finer‑grained view than broader genomic predictors. Researchers say this granularity can expose disease‑relevant effects that broader models may miss. The development marks a significant step toward linking human genetics with cellular biology.
AlphaGenome 2 was trained on more than 100 million individual cell profiles spanning 68 tissues, converting static cell atlases into a predictive engine for human genetics. According to DeepMind, the model identifies nearly twice as many variants that alter gene expression compared with its predecessor. It also demonstrates the ability to predict outcomes for novel cell types and states without retraining, filling gaps in experimental data. This capability can guide laboratory investigations toward the most informative experiments. The extensive training set underpins the model’s unprecedented resolution.
Higher resolution uncovers disease‑relevant genetic effects
DeepMind scientists highlight that AlphaGenome 2 can pinpoint variants that affect gene activity in disease‑associated cells, a task where broader models often fall short. The system’s detection rate of expression‑changing variants is almost double that of the original AlphaGenome. Moreover, it can extrapolate predictions to previously uncharacterised cellular states, offering insights into unexplored biological contexts. Such precision is valuable for drug developers seeking targets within specific tissues or cell populations. Adam Gayoso, a research scientist on the project, described the achievement as “tying human genetics to cell biology.”
Genomics lead Žiga Avsec emphasized that drug discovery relies on targeting particular proteins in defined cell types, making AlphaGenome 2’s granularity essential. When a variant’s impact is mapped to a specific cell, therapeutic strategies can be more accurately tailored. Avsec noted that recent advances in single‑cell technologies have finally provided the data volume required for models of this scale. The team plans to expand training data to under‑represented cell states and eventually incorporate multi‑species datasets. In the longer term, AlphaGenome 2 could feed into “virtual cell” simulations that model complex biological systems.
API launch, computational cost and future roadmap
Google DeepMind is making AlphaGenome 2 accessible through a non‑commercial research API starting today, with a limited private preview for commercial projects. The company cautioned that the model’s increased predictive detail comes with higher computational expense, rendering it impractical for generating a genome‑wide atlas of every possible DNA change. Consequently, AlphaGenome 2 will complement rather than replace the original AlphaGenome for large‑scale reference maps. The API rollout enables researchers to harness the model without building dedicated infrastructure. DeepMind is already working on the next generation, aiming to enrich data coverage and explore cross‑species training.
Looking ahead, AlphaGenome 2 is expected to become a core component of digital “virtual cells,” which simulate biological processes for research and drug discovery. Its release aligns with a broader push by Google and rival AI labs to apply artificial intelligence to major scientific challenges, from materials science to drug design. DeepMind has hinted at opening an automated materials‑science laboratory in the UK later this year. AlphaGenome 2’s ability to predict cellular effects of genetic variation positions it as a pivotal tool in precision medicine and future AI‑driven biology initiatives.
TechTimes analysis
TechTimes believes AlphaGenome 2 represents a notable advance in the convergence of AI and genomics, as its cell‑type specific predictions enable more accurate identification of disease‑linked variants. While the higher computational cost limits its use for massive genome‑wide projects, the newly released API gives researchers immediate access for focused studies. In an increasingly competitive landscape of AI‑driven biomedical research, AlphaGenome 2 could become a critical asset for drug target discovery and mechanistic disease understanding, while also laying groundwork for future virtual‑cell simulations.
TechTimes analysis
TechTimes believes AlphaGenome 2 represents a notable advance in the convergence of AI and genomics, as its cell‑type specific predictions enable more accurate identification of disease‑linked variants. While the higher computational cost limits its use for massive genome‑wide projects, the newly released API gives researchers immediate access for focused studies. In an increasingly competitive landscape of AI‑driven biomedical research, AlphaGenome 2 could become a critical asset for drug target discovery and mechanistic disease understanding, while also laying groundwork for future virtual‑cell simulations.