New Google AI tool can study DNA changes and predict disease risk

Google DeepMind's AlphaGenome can analyse long stretches of DNA to predict how hidden genetic changes affect gene regulation, offering new clues to cancer, heart disease and mental illness

Scientists may now be closer to understanding how hidden changes in our DNA lead to disease, thanks to a new artificial intelligence tool from Google DeepMind.

 

Detailed in the study Advancing regulatory variant effect prediction with AlphaGenome, published in the journal Nature, the tool goes far beyond traditional genetic analysis. It can scan up to one million DNA “letters” at once and predict how tiny genetic changes influence when, where and how strongly genes are switched on, often the hidden drivers of disease.

 

Researchers say this approach could help explain why conditions such as cancer, heart disease, and mental health disorders develop, by shedding light on parts of the genome that have long been difficult to interpret.

 

What is AlphaGenome and why is it being called a breakthrough?

AlphaGenome is a deep-learning model designed to understand gene regulation, the complex control system that tells genes when to act and when to stay silent. Unlike earlier tools that analysed only short DNA fragments, AlphaGenome can process very long stretches of DNA while still spotting changes at single-letter precision.

 

This matters because gene regulation often depends on distant DNA elements working together. By capturing this long-range context, AlphaGenome offers a more realistic picture of how the genome functions inside living cells.

Why do most disease-linked mutations sit outside genes?

Only about two per cent of the human genome directly codes for proteins. The rest acts like a vast control panel, regulating gene behaviour.

 

Many conditions, including heart disease, autoimmune disorders, psychiatric illnesses, and cancers, are linked not to broken genes but to faulty regulation. Until now, identifying which regulatory mutations actually cause harm has been extremely difficult. AlphaGenome is built precisely to tackle this problem.

How does AlphaGenome read DNA differently from earlier AI models?

According to the Nature study, earlier AI models could either analyse long DNA sequences with low detail or short sequences with high precision. AlphaGenome breaks this compromise.

 

It predicts thousands of molecular readouts, such as gene expression levels, splicing patterns, chromatin accessibility and even three-dimensional DNA folding, at near base-pair resolution. This unified view allows scientists to see how a single mutation can ripple across multiple biological processes.

What did the study find about AlphaGenome’s accuracy?

In extensive testing, AlphaGenome matched or outperformed the best existing models in 25 out of 26 benchmarks for predicting the effects of genetic variants.

 

Crucially, it accurately reproduced known disease mechanisms, including cancer-related regulatory changes near the TAL1 oncogene, a gene implicated in T-cell leukaemia. This gives researchers confidence that the model is capturing real biology, not just statistical patterns.

 

According to the study, AlphaGenome can pinpoint which non-coding mutations are likely to drive these changes.

 

The tool also showed strong performance in predicting splicing errors, a common cause of rare genetic diseases, by modelling how mutations disrupt the way RNA is stitched together inside cells. 

ALSO READ: ChatGPT Health promises personalised care, but it comes with many risks

Can AlphaGenome replace laboratory experiments?

While the study suggests AlphaGenome can dramatically narrow down which mutations are worth testing in the lab, saving time, money and effort, the authors caution that its predictions still need experimental validation because human cells do not always behave exactly as models expect.

 

However, by predicting how specific DNA sequences activate genes in particular cell types, AlphaGenome could help scientists design more precise gene therapies.

 

It may also improve genetic risk prediction by clarifying which variants truly matter, an important step towards personalised medicine based on biological function rather than statistical association alone.