AlphaGenome: Google DeepMind’s Breakthrough AI for Decoding DNA Function
Google DeepMind launched AlphaGenome in June 2025. The team published the full research in Nature on January 28, 2026. This AI model represents a major advance in genomics. Researchers compare it to AlphaFold, which revolutionized protein structure prediction.
AlphaGenome acts as a unified DNA sequence model. It takes long DNA inputs—up to 1 million base pairs (1 Mb). The model processes this sequence in one go. Then, it predicts thousands of functional genomic tracks. These predictions reach single-base-pair resolution for most outputs.
The model focuses on regulatory elements. These elements control gene activity. Most lie in non-coding DNA, often called the “dark matter” of the genome. Non-coding regions make up about 98% of human DNA. They do not code for proteins. Instead, they regulate when, where, and how much genes express.
AlphaGenome predicts diverse molecular phenotypes. Key outputs include:
- Gene expression levels
- Splicing patterns
- Chromatin accessibility
- Histone modifications
- Transcription factor binding
- 3D genome folding and contact maps
The model excels at variant effect prediction. It analyzes how single mutations (variants) alter these functions. For example, a tiny change can disrupt enhancer activity. This may switch a gene on or off at the wrong time or place. Such disruptions link to diseases like cancer, dementia, obesity, and rare genetic disorders.
DeepMind
trained AlphaGenome on vast datasets from human and mouse genomes. The model builds on earlier work like Enformer. It outperforms previous tools in most benchmarks. In 25 out of 26 variant-effect evaluations, AlphaGenome matches or beats the best external models.
Scientists access AlphaGenome easily. An API lets researchers run predictions programmatically. A free version supports non-commercial use. The model code is available on GitHub. Over 3,000 scientists from 160 countries have already tested it since the initial release.
This tool accelerates discovery. It helps interpret genetic variants from genome-wide association studies (GWAS). Many disease-linked variants hide in non-coding regions.
AlphaGenome connects them to real biological effects. As a result, it aids precision medicine, drug target identification, and rare disease diagnosis.
Experts call AlphaGenome a “Swiss-army knife” for non-coding DNA. It deciphers the complex regulatory code.
Ultimately, it promises deeper understanding of the “code of life” and new paths to treatments. The field moves fast—expect more applications soon.