Researchers at DeepMind, a subsidiary of Google specializing in artificial intelligence (AI), have unveiled a groundbreaking tool that predicts the potential pathogenicity of genetic mutations, offering significant potential for rare disease research.
This innovative tool is designed to assess “missense mutations”, genetic alterations characterized by a single nucleotide change, akin to a typographical error in the DNA code. Such mutations can result in changes to the resulting amino acid sequence, impacting health. While an individual typically carries around 9,000 of these mutations, the majority are benign. However, some are responsible for severe diseases like cystic fibrosis, sickle-cell anemia, or cancer.
Presently, there are approximately four million documented missense mutations in humans, yet only a mere two percent have been definitively classified as pathogenic or benign. This leaves a vast number, around 71 million, without clear categorization. To address this challenge, DeepMind developed the AlphaMissense tool, which successfully categorized 89 percent of these mutations.
How Does AlphaMissense Work? AlphaMissense assigns each mutation a score ranging from 0 to 1, indicating its potential pathogenicity or disease-causing capacity. It estimates that 57 percent of these mutations are likely benign, while 32 percent are likely pathogenic, leaving the rest in an uncertain category. The comprehensive database generated by AlphaMissense is publicly accessible to scientists via GitHub, a platform for storing and sharing computer code. A study detailing these findings was published in the journal Science.
Experts Joseph Marsh and Sarah Teichmann noted in their article in Science that AlphaMissense demonstrated superior performance, highlighting its potential impact.
AlphaMissense’s training is based on an extensive database of human and primate DNA, enabling it to identify prevalent genetic mutations accurately. According to Jun Cheng, a scientist at Google DeepMind, the tool assesses whether a protein sequence is concerning in terms of potential health risks. The predictions generated by AlphaMissense could significantly accelerate rare disease diagnosis and potentially lead to the discovery of new disease-associated genes. While researchers anticipate that this could indirectly pave the way for new treatments, they emphasize that AlphaMissense should not be used as the sole diagnostic tool.
AlphaMissense’s development is rooted in AlphaFold, another Google DeepMind machine learning program introduced in 2018. AlphaFold pioneered the creation of the largest protein database with over 200 million available structures, demonstrating the company’s commitment to advancing bioinformatics and genetics research.
