Google has announced a significant advancement in identifying disease-causing genes using artificial intelligence (AI). The model, called AlphaMissense, is capable of accurately classifying 89% of missense variants in genes, determining whether they are likely to cause diseases or are benign. In comparison, human experts are only able to confidently classify 0.1% of such variants.
Missense variants occur when a single letter in the DNA is substituted, resulting in proteins with different amino acids. While most of these variants are harmless, some can lead to rare genetic disorders. The AlphaMissense model analyzes existing information on missense variants, their commonality in humans, and closely related primates. By focusing on rarely seen variants, the model can classify them as pathogenic and use that information to assess other protein sequences. It not only provides a verdict on their potential to cause problems but also indicates the confidence level of the prediction.
Traditionally, identifying disease-causing mutations through human experiments is expensive and time-consuming, requiring the examination of each unique protein separately. Google’s new system allows researchers to preview results for thousands of proteins simultaneously, assisting them in determining areas of focus. The company has also released a vast catalogue of missense mutations, enabling researchers to understand their effects. This knowledge is crucial for studying and developing treatments for conditions like cystic fibrosis, sickle-cell anemia, and cancer.
Google’s Deepmind division, known for its work in artificial intelligence, continues to make strides in healthcare. AlphaMissense is built on AlphaFold, the breakthrough AI model used to predict protein folding. The research and development behind the system are detailed in a paper titled ‘Accurate proteome-wide missense variant effect prediction with AlphaMissense’, published in the journal Science. Google has made the catalogue freely available to the research community, along with the code for the AI system.
Sources:
– ‘Accurate proteome-wide missense variant effect prediction with AlphaMissense’, Science.
