AlphaGenome 2 reportedly adds single-cell predictions, with striking results still to test

Google DeepMind has reportedly released AlphaGenome 2, an AI model that predicts how genetic variants affect different cell types. 36Kr reports near-doubling on two benchmark measures, a potentially useful advance for researchers studying disease-linked DNA that does not code for proteins.

Google DeepMind Watch analysis

What happened

The report says AlphaGenome 2 can make predictions at single-cell level, rather than averaging signals across whole tissues. It gives benchmark results in which the model identified disease-causing variants in 30 per cent of cases, up from 16 per cent for a fine-tuned first-generation model. Accuracy in predicting whether a variant raises or lowers gene activity reportedly rose from 23 to 43 per cent.

36Kr also says DeepMind has made variant-impact data and research tools available, and describes tests involving Parkinson’s disease samples. These are claims in the publication’s account; the supplied evidence does not include DeepMind’s announcement or the underlying paper.

Why it matters

A signal that differs between cell types can disappear in a tissue-wide average. If the reported results hold up, finer-grained predictions could help researchers decide which genetic variants and cell types merit laboratory investigation. That is a research aid, not a diagnosis or a substitute for experiments, however much the headline numbers might tempt us to skip ahead.

Our read

The reported benchmark gains are worth attention, especially alongside the shift to single-cell predictions. But the article’s bold claims need the paper and methods behind them before readers can judge what was tested and how broadly the findings apply. Interesting lead, not a medical breakthrough to bank just yet.

What to watch

• The paper and its methods, including how the benchmarks were constructed.
• Whether the reported datasets and tools are available to researchers as described.
• Independent evaluation of the results on new samples and experiments.

Discussion spark: If an AI model can narrow down which genetic variants deserve lab work, what evidence should researchers require before relying on its predictions?

not affiliated with, endorsed by, or operated by Google or Google DeepMind