Google DeepMind: New AI tool predicts genetic diseases

a new program Artificial intelligence developed by scientists at Google DeepMind It can predict whether there are millions of genetic mutations in our DNA sequence that may cause diseases in the future. The tool, called AlphaMissense, uses powerful learning algorithms to assess the pathogenicity of so-called “missense variants.”

Inaccuracies are genetic mutations in which a single letter (nucleotide) in the DNA code is replaced by another letter. By coding for a different amino acid in a protein sequence, It can either function normally (even if its efficiency is reduced) or its function can be altered, making it a non-functional protein.

Although not all missense mutations cause major protein changes, the lack of accurate functional predictions of their effects “limits the diagnostic rate of rare diseases,” he says. These diseases associated with genetic changes range from cystic fibrosis and sickle cell anemia to cancer and brain developmental disorders.

ordinary person There are around 9 thousand missense mutations in its genome. However, of the more than 4 million cases observed, only 2% were classified as benign or pathogenic. Doctors already have computer programs for such predictions, but they are inaccurate and work only for diagnostic support.

Powered by DNA data from humans and monkeys, AlphaMissense learns which missense mutations are common/benign and which are rare and potentially dangerous. In this “training” the program became familiar enough with the language of proteins to know how a “healthy” macromolecule behaves.

Artificial Intelligence previews degenerative diseases

A review by Google reveals the following about its artificial intelligence model AlphaMissense was able to classify 89% of the 71 million possible missense variants. “The idea is, instead of giving me a change in a protein and predicting the shape of the protein, I tell you: Is this bad for the human who has it?” physicist Stephen Hsu explains to Interesting Engineering. , from Michigan State University.

Big question regarding amino acid substitutions for the expert who evaluated genetic problems with AI techniques who was not involved in this study Not having the slightest idea whether they cause diseases or not. However, predicting the effects of a particular variable is an expensive and complex process.

By combining “structural context and evolutionary conservation,” AlphaMissense achieves state-of-the-art results across a wide range of genetic benchmarks, according to the paper’s authors. “all without explicit training on such data”. This implies the simultaneous preview of thousands of proteins that can “prioritize resources and speed up more complex studies,” the study concluded.

Keep up to date with science and technology (or any other notebook) on TecMundo. If you want, take the opportunity to read about the quantum method that can treat brain cancer using spray.