DeepMind, an artificial intelligence firm owned by Google, has adapted its AlphaFold system to predict the potential harmfulness of DNA mutations in proteins. The adapted system, called AlphaMissense, focuses on missense mutations, where one of the DNA letters in a triplet is changed to another letter, potentially impacting the functionality of the protein. AlphaMissense has analyzed 71 million possible missense mutations in 20,000 human proteins. The results of the analysis have been made freely available to clinicians and human geneticists, providing them with assistance in pinpointing the cause of genetic diseases. This innovative approach holds promise in helping doctors accurately diagnose diseases by identifying the specific mutations that are linked to a particular condition.
The AlphaMissense system compares the sequence of each possible mutated protein to those stored in its database, trained by AlphaFold, to determine if the mutation appears “natural”. Proteins that are identified as “unnatural” are rated as potentially harmful. While this approach does not assess the structural or stability changes in the protein caused by the mutation, it performs better than other computational methods in identifying known variants. The AlphaMissense system has demonstrated remarkable results and will be particularly valuable in prioritizing disease-causing mutations for further investigation.
It is important to note that missense mutations are just one type of mutation, and there are other genetic alterations such as additions, deletions, duplications, and rearrangements that can also contribute to diseases. Additionally, some disease-causing mutations do not directly affect proteins but instead impact gene regulation. Nevertheless, this AI-based system serves as a valuable tool in aiding the diagnosis process for genetic diseases.
