AlphaGenome Atlas: An AI Map of Every Possible DNA Change

Google DeepMind has launched a new online platform called AlphaGenome Atlas. It's a massive database of AI predictions covering every possible single-letter change in human DNA — more than 9 billion of them. Testing each one in a lab would be practically impossible; there are simply too many.

The Atlas is built on AlphaGenome, DeepMind's AI model that predicts how genetic variants affect biological processes such as gene regulation, protein structure, and splicing (how pieces of a gene get joined together). The dataset is 1 petabyte in size — equivalent to 1,000 terabytes — more than 30 times larger than the earlier AlphaFold database.

Inside the Atlas is a new score called AVI (AlphaGenome Variant Impact), which combines results from AlphaGenome and another model, AlphaMissense, into a single number showing how harmful a given DNA change might be. The score also shows what's driving the assessment — for example, whether the issue involves chromatin structure, splicing, or gene expression.

Rare-disease researchers and geneticists have already used the tool to identify DNA changes linked to certain diseases, and have confirmed some of these findings through lab experiments.

The tool is free for academic and research use, accessible via a website, an API, and as a feature in Google Antigravity. This news does not cover clinical use — like testing patients directly — that hasn't been addressed yet. There are also no full accuracy figures for the AVI scores published so far.

For ordinary Tanzanians, this isn't a tool you'd use directly. But for biology students, science teachers, and health journalists, it's an important resource for following how genomics research is advancing worldwide.