Researchers at Google DeepMind have developed an artificial intelligence (AI) tool called AlphaMissense, which can predict whether mutations in human genes are likely to be harmful. This technology has the potential to accelerate the diagnosis of diseases caused by genetic variants. The AI tool analyzed 71 million “missense” mutations, where a single letter in the human genetic code changes. Of these mutations, 32 percent were classified as likely to be pathogenic, 57 percent as benign, and the remaining uncertain.
The findings of this research have been published in the journal Science. Currently, human experts have only been able to identify the clinical effect of 0.1 percent of these variants. Discovering disease-causing mutations is expensive and time-consuming, as experiments need to be designed separately for each protein in the body. However, by using AI predictions, researchers can preview results for thousands of proteins simultaneously, thus aiding in resource prioritization and accelerating complex studies.
It is important to note that the predictions made by AlphaMissense should not be used as the sole basis for clinical diagnosis. They should be utilized in conjunction with other evidence. Nonetheless, these predictions have the potential to increase the diagnosis rate of rare diseases and contribute to the identification of new disease-causing genes.
Genomics England, a UK government initiative, tested the tool’s predictions against its own extensive records of genetic variants causing rare diseases and was impressed by the results. The tool is expected to be used in healthcare as a co-pilot for clinical scientists, helping to prioritize variants and enhance efficiency in their work.
AlphaMissense is an extension of DeepMind’s AlphaFold tool, which predicts protein structures. The AI tool incorporates a vast amount of biological evidence regarding the characteristics of mutations in humans and other primates.
The tool has been made freely available to the scientific community and its predictions will be integrated into the Ensembl Variant Effect Predictor, widely used by researchers in the field.
While AlphaMissense is a significant advancement, it has limitations. Its predictions are made in a general sense and do not provide detailed information about the biophysical nature of a variant. Further development of the tool may reveal more insights in the future.
Overall, this AI technology has the potential to revolutionize the field of genetic research and contribute to the understanding and diagnosis of diseases caused by genetic mutations.