DeepMind's AlphaMissense AI tool helps classify genetic mutations

DeepMind announced on Tuesday that a study published in the journal Science showed that its new artificial intelligence (AI) model, called AlphaMissense, could classify the effects of missense mutations, in which a single letter of the genetic code changes. The company said it has made the AlphaMissense catalogue of mutations freely available to the scientific community and open sourced the model code for AlphaMissense.

"Experiments to uncover disease-causing mutations are expensive and laborious," said author Žiga Avsec, adding that "by using AI predictions, researchers can get a preview of results for thousands of proteins at a time." Co-author Jun Cheng said that "the predictions were never really intended to be used for clinical diagnosis alone. They should always be used along with other evidence." However, Cheng added that "our predictions will help to increase the diagnosis rate of rare disease and also potentially to help us find new disease-causing genes."

In the study, the tool categorised 89% of all 71 million possible missense variants, predicting 57% were likely benign and 32% were likely pathogenic. DeepMind noted that, by contrast, only 0.1% of the variants have been confirmed by human experts. Avsec noted that the AI tool's predictions of pathogenicity "are made in a general sense and do not tell us the biophysical nature of what a variant does," adding that such insights might emerge as the tool is further developed.

Built on AlphaFold tool

The AI tool was built on DeepMind's AlphaFold algorithm, which predicts protein structure. The tool also learnt from a vast amount of biological evidence about the characteristics of mutations in humans and primates that make a genetic variant pathogenic or benign. AlphaMissense makes predictions across a wide range of genetic benchmarks without training on such data, DeepMind said. It "leverages databases of related protein sequences and structural context of variants to produce a score between 0 and 1, approximately rating the likelihood of a variant being pathogenic,' the company explained. 

"Our tool outperformed other computational methods when used to classify variants from ClinVar, a public archive of data on the relationship between human variants and disease," DeepMind said. The company added that the AI tool was "the most accurate method for predicting results from the lab, which shows it is consistent with different ways of measuring pathogenicity."

Genomics England cross-reference

DeepMind has been working with Genomics England to explore how these predictions could help study the genetics of rare diseases. Genomics England cross-referenced AlphaMissense's findings with variant pathogenicity data previously aggregated with participants and confirmed that the predictions were accurate and consistent.

Ellen Thomas, deputy chief medical officer at Genomics England, said that the NHS will be among the first organisations to benefit from the new development. "The new tool is really bringing a new perspective to the data. It will help clinical scientists make sense of genetic data so that it is useful for patients and for their clinical teams," she said.

Don’t want to miss our top stories? Sign up for our free daily newsletter here.