Google's AlphaGenome evaluates all one-base genome changes

The human genome is mostly non-coding DNA, much of which appears to be junk from ancient infections or inactivated genes. Biologists have struggled to identify which non-coding regions are functional, partly because DNA-binding proteins are not very specific and often work in clusters. This makes it hard to predict the impact of mutations in these areas.

Google's AlphaGenome applies AI to this problem, trained on limited but well-studied cell types. For researchers, it offers a way to quickly assess whether changes in non-coding regions of a gene are likely to matter, along with a hypothesis for why. Its predictions match or exceed specialized tools, but it doesn't give definitive answers.

The real test will be whether AlphaGenome's hints translate into successful experimental validations. Since it hasn't been trained on all cell types or organisms, its reliability may vary beyond well-characterized conditions. Wider adoption may depend on expanding training data to more diverse biological contexts.