Researchers at Google DeepMind, the US giant’s artificial intelligence company, unveiled a tool on Tuesday that predicts whether genetic mutations are potentially pathogenic or not, a discovery that could particularly help research into rare diseases, AFP reported.

It’s “a new step in recognizing the impact of artificial intelligence on the natural sciences,” Ziga Avsek, vice president of Google DeepMind research, said at a press conference.

The tool analyzed so-called “missense” mutations, where one segment of the genetic code is affected, resulting in a change in protein function. Each cell performs its function with the help of proteins that constantly transmit its instructions.

Humans have about 9,000 such mutations, most of which are benign, according to the company, but some can lead to diseases such as cystic fibrosis.

To date, four million of these mutations have been identified in humans, but only 2% of them have been classified as either pathogenic or benign.

In total, there are 71 million possible mutations of this type. They were tested by Google’s DeepMind AlphaMissense tool, which was able to pronounce 89% of them.

Each mutation received a score from 0 to 1, indicating the risk that it is pathogenic, that is, causes disease. Result: 57% were classified as probably benign and 32% as probably pathogenic, the rest were indeterminate.

The database was made public and became available to all scientists. A study that proves this was published on Tuesday in the prestigious journal Science.

“High productivity

Experts Joseph Marsh and Sarah Teichmann write in a paper published in Science that AlphaMissense demonstrates “superior performance” compared to existing tools.

The tool “doesn’t say what disease it will cause,” Google DeepMind researcher Jun Cheng explained during a press conference. “But we believe that our predictions can be useful for improving the diagnosis of rare diseases, and potentially help us find new genes associated with diseases.”

Indirectly, this could lead to the development of new treatments, say the researchers, who caution, however, that AlphaMissense should not be used alone to make a diagnosis.

The tool was trained on a database of human and primate DNA, allowing it to recognize which genetic mutations are common.

“It learns what a protein sequence normally looks like, and when it’s presented with a sequence with a mutation, it can tell if it looks worrisome or not,” Jun Cheng explained.

“It’s a bit like human language: if you change a word in an English sentence, a person who knows English can immediately see whether the change affects the meaning of the sentence or not,” he compared.

AlphaMissense was based on AlphaFold, another machine learning program introduced more than two years ago by Google DeepMind, which made it possible to publish the largest database of proteins.