Signal

Google DeepMind launches AlphaGenome 2, trained on 100M+ cell profiles across 68 tissues to predict genetic mutation effects, for non-commercial research

First reported by The Verge ·

The signal ●●●● Compiled by AI from The Verge, Techmeme and Google DeepMind
Why you might care

Researchers can now directly investigate the specific cellular impact of genetic variations, accelerating the discovery of disease mechanisms and targeted treatments.

What happened

Google DeepMind has launched AlphaGenome 2, an advanced AI model designed to predict the effects of genetic mutations on individual cell types and states. This successor to last year's AlphaGenome was trained on over 100 million cell profiles from 68 tissues, offering a significantly finer-grained view of genetic variant impacts than its predecessor. The model can identify nearly twice as many variants that alter gene expression and, crucially, make predictions for new cell types and biological states without retraining. AlphaGenome 2 aims to bridge the gap between human genetics and cell biology, enabling researchers to pinpoint cellular-level effects of genetic variations and potentially link them to diseases. The tool is now available for non-commercial research via an API, with a limited private preview for commercial entities.

What it means

AlphaGenome 2's ability to predict effects across untrained cell types and states signals a leap in AI's capacity for biological inference, moving beyond simple pattern matching to a more generalized understanding of genetic regulation. This could dramatically reduce the time and cost of experimental validation, allowing scientists to prioritize investigations with higher certainty of relevance. The model's potential to identify experimental data gaps further streamlines the research pipeline.

The release, particularly its availability for non-commercial research, democratizes access to sophisticated genomic analysis tools previously limited to large institutions. This empowers a broader scientific community to explore complex genetic questions, potentially leading to novel therapeutic strategies and a deeper understanding of diseases. The ongoing development hints at future AI models capable of cross-species analysis and virtual biological simulations.

AI-written summary. May contain errors.