AlphaGenome takes in up to 1 million DNA base pairs and predicts thousands of functional genomic readouts at single‑base resolution, across 11 assay modalities including gene expression, chromatin accessibility, histone marks, transcription factor binding and 3D chromatin contacts (Agarwal et al., Nature, 2026). This unifies what has typically required multiple specialized models and provides both long‑range context and nucleotide‑level detail for regulatory variant effect prediction.

In benchmarking, the model matched or outperformed state‑of‑the‑art tools in 25 of 26 tasks, including eQTL effects, enhancer–gene interactions, alternative polyadenylation and splicing (Agarwal et al., Nature, 2026). Performance gains were particularly strong for noncoding variants, which lie in the 98% of the genome that does not code for proteins but is increasingly linked to complex disease risk.

Implications for life science R&D

For industry, AlphaGenome’s main value lies in rapid in silico functional annotation of variants and regulatory elements. Potential applications include prioritising noncoding variants in rare disease and cancer, identifying regulatory drivers and designing cis‑regulatory sequences for gene and cell therapies or RNA‑based drugs (BioWorld, 2026; Williams, GEN, 2026). Because the model can score the impact of a single variant across thousands of molecular features in about a second, it could help triage which hypotheses to take into wet‑lab validation, shortening cycles in functional genomics and target validation (Williams, GEN, 2026).

DeepMind has released AlphaGenome for non‑commercial use, opening the door for academic and translational groups – to integrate the model into existing variant‑interpretation and functional genomics pipelines.

Independent experts note that clinical use will require careful validation across diverse populations and disease contexts, but broadly view AlphaGenome as a step change for regulatory genomics, analogous to AlphaFold’s impact on protein structure prediction (Science Media Centre, 2026). DeepMind has released AlphaGenome for non‑commercial use, opening the door for academic and translational groups – to integrate the model into existing variant‑interpretation and functional genomics pipelines (Agarwal et al., Nature, 2026).