New tool from Google's AI lab hunts for genetic disease causes
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Maura Losch/Axios
A new program from Google's artificial intelligence lab DeepMind aims to tackle one of the toughest problems in genetics: sifting through the millions of variations in the human genome to predict those that ultimately cause disease.
Why it matters: The effort could help researchers more quickly zero in on the genes responsible for diseases — especially for rare ones — and identify promising new targets for drug development. But similar tools haven't been widely used, and it would need extensive testing before it could be deployed in medical practice.
Driving the news: DeepMind officials say the new AI tool called AlphaMissense, announced Tuesday in the journal Science, can identify certain mutations in the genome that were previously unknown to cause disease.
- The company is releasing a free database of those predictions for all possible single-letter mutations in the human DNA code, classifying 89% of what are known as "missense" variants as either likely benign or likely pathogenic.
Details: A gene's DNA sequence can be altered in various ways, including what are known as "missense" variants, which may or may not affect a person's health.
- There are millions of potential variations, making it difficult for researchers to parse all of them.
- It's "really a needle-in-the-haystack problem," Žiga Avsec, a DeepMind research scientist, told reporters this week.
- Researchers from DeepMind said the AlphaMissense algorithm was able to make predictions among 71 million missense variants about which might lead to disease.
- Applying the algorithm to classify those variants, it was able to predict 57% of the variants were likely benign and 32% were likely disease-causing.
- Using data about DNA from humans and some primates to learn the "language" of proteins, the algorithm produces a score rating the likelihood that a variant could be disease-causing.
- "This is very similar to human language," said Jun Cheng, research scientist and team lead at DeepMind. "Like if we substitute a word from an English sentence, a person that is familiar with English can immediately see whether this word substitution will change the meaning of the sentence or not."
Reality check: This builds off work from DeepMind's AlphaFold, an AI tool that can predict protein structures for the entire human genome.
- While AlphaFold ushered in a new era in computational biology, this tool isn't the same leap forward, Nature reported.
- "It's exciting. It's probably the best predictor we have right now. But will it be the best predictor in two or three years? There's a good chance it won't be," Joseph Marsh, a computational biologist at the MRC Human Genetics Unit in Edinburgh, told Nature.
- Researchers also cautioned the Al's predictions need to be verified by other scientists.
- "A model like this may turn out to be more complicated than the biology it is trying to predict," Ben Lehner, a human genetics expert at the Wellcome Sanger Institute, told The Guardian.
The DeepMind team noted the tool is not meant to be used for clinical diagnosis alone.
- "However, we do think that our predictions can potentially be helpful to increase the diagnosis rate of rare disease, and also potentially to help us find new disease-causing genes."
