AlphaGenome 2: Teaching AI to read the entire human genome

AlphaGenome 2: Teaching AI to read the entire human genome

More

Summary

Google DeepMind’s podcast explores AlphaGenome 2, the latest model designed to help scientists read the human genome. Sequencing human DNA took 13 years and produced a text far longer than the complete works of Shakespeare, but understanding what it does has taken far longer. Two DeepMind researchers explain how deep learning is making that book more legible.

AlphaGenome reads a stretch of DNA and predicts what is happening in the cell at that location, including what changes if a single letter is altered. The first version, unveiled in 2025, predicted the function of non-coding DNA across whole tissues. AlphaGenome 2 zooms in to individual cell types to show how single mutations may drive disease. The researchers describe their backgrounds, explain genotype versus phenotype, and discuss why connecting the two is so hard.

A highlight is a case from the Broad Institute, where researcher Laura Koval applied the variant scoring to an infant with a rare epilepsy whose causal mutation had gone unidentified for years. The top-ranked variant matched the condition’s biology. The team stresses that this is a research tool for prioritizing variants within a larger diagnostic pipeline, not a clinical product.


๐Ÿ“บ Source: Google DeepMind ยท Published November 11, 2026
๐Ÿท๏ธ Format: Podcast

1 Item

Channels

1 Item

Companies