Google AI system predicts effects of every single-base genome change

The primary focus of AlphaGenome is on non-coding DNA, which, despite not encoding proteins, comprises the vast majority of the genome and is crucial for regulating gene expression. This includes sequences that dictate when and where genes are activated, how messenger RNAs are processed, and how DNA is packaged within cells.

While much of the non-coding DNA is considered 'junk'—remnants of viral infections or inactivated genes—identifying functional sequences remains a significant challenge for biologists. The AlphaGenome AI system is designed to tackle this by analyzing probabilities and contextual information, aiming to identify gene expression, transcription initiation, and chromatin accessibility, among other factors.

Currently, the system's capabilities are limited to sequences from mice and humans, and it has been trained on a finite number of cell types. However, it can assist researchers by providing insights into the potential significance of mutations found in non-coding regions and offering hypotheses for their effects. Its predictions are reported to be comparable to or better than existing specialized software tools.

Google's motivation for undertaking this comprehensive analysis includes pre-calculating potential mutations of interest for researchers, allowing for immediate assessment of newly discovered mutations. It also enables scientists to search the entire genome for specific types of impact. The long-term potential lies in extending the system's analytical capabilities to genomes for which training data is scarce, such as Neanderthal or Denisovan genomes, or cell types not yet exhaustively studied.