Scientists from Harvard Medical School in the United States and the Genome Regulation Center in Barcelona, Spain, published research results in the journal Nature Genetics on the 24th, stating that they have developed an artificial intelligence (AI) model called popEVE. This model can accurately identify the mutations in human proteins that are most likely to cause diseases, and this breakthrough is expected to completely change the diagnosis of genetic diseases. The development of popEVE is based on evolutionary data from hundreds of thousands of different species and genetic variation information from the entire human population. The vast evolutionary record enables the tool to decipher the key and variable regions of over 20000 human proteins, enabling it to not only identify pathogenic mutations but also rank the degree of harm caused by these mutations to the human body. Each person's genome contains many small variations, including "missense mutations" that alter individual amino acids in proteins. Most of these mutations are harmless, but some may cause serious illnesses. The real challenge lies in how to distinguish between benign and harmful mutations. In addition, the degree of impact of harmful mutations varies: some only cause mild symptoms, some lead to severe disabilities, and some even endanger life in childhood. Most existing AI tools can only predict whether mutations are dangerous, but it is difficult to assess their degree of harm. What's even more tricky is that 'rare mutations' lack case references, and even with global genetic sequencing, these mutations are often untraceable, leaving traditional methods relying on patient population data powerless to address them. PopEVE innovatively combines evolutionary data with two major resource repositories, the UK Biobank and Genome Aggregation Database. By analyzing the genetic variations present in the healthy population, the model was able to calibrate its predictions for human diseases. This enables scientists to establish for the first time a model that can rank the hazards of mutations in the entire proteome, which can help doctors prioritize the most destructive mutations. To validate the effectiveness of the model, the team analyzed genetic data from over 31000 families of children with severe developmental disorders. In 98% of cases, popEVE marks common mutations as the most destructive variants, outperforming similar advanced tools including AlphaMissine, a company specializing in deep thinking. Moreover, in exploring new pathogenic genes, popEVE discovered 123 genes previously thought to be unrelated to developmental disorders, of which 104 only appeared in individual cases. (New Society)