New tool facilitates clinical interpretation of genetic information

In recent years, genomic sequencing has become less expensive and more advanced. On the one hand, this allows clinicians to increasingly use sequencing for diagnostic purposes while also allowing scientists to explore more research hypotheses. On the other hand, many detected mutations do not have a clear clinical interpretation. The uncertainty over whether a mutation causes disease can be stressful for patients and lead to psychological burden, morbidity and health-care expenses associated with under- and overdiagnosis. While existing tools are already used to predict the functional impact of these variants, their performance is biased due to limited clinical data that makes distinguishing between pathogenic (disease-causing) and benign (neutral) variants within a given gene difficult and often lead to misclassifying mutations that do not cause disease as pathogenic. Addressing these difficulties is critical for developing a reliable predictor for clinical applications.

The research group of Agnes Toth-Petroczy at the MPI-CBG and the CSBD teamed up with Christopher Cassa, assistant professor of medicine at the Brigham and Women's Hospital Division of Genetics at the Harvard Medical School, and Ivan Adzhubei, research associate at the Department of Biomedical Informatics at Harvard Medical School, to develop a statistical model and web server DeMAG that reaches high accuracy in the interpretation of genetic mutations in disease genes. To do this, the researchers carefully selected known pathogenic and benign mutations for training the model. "We used clinical and various population databases. We selected only mutations whose clinical interpretation is agreed upon among multiple submitters such as medical doctors and genetics laboratories. And we also included data from ancestries that are underrepresented in the current population databases, such as Korean or Japanese, to make it even more representative and accurate," explains Federica Luppino, first author of the research paper and PhD student in the Toth-Petroczy group. DeMAG includes a novel feature, the "partners score," that identifies clusters of amino acids in a protein that share the same clinical effect. With the partners score, DeMAG takes advantage of the amino acid relationships based on evolutionary information from the genomes of many organisms and the recent AI (Artificial Intelligence) revolution of predicting the 3D shapes of proteins using the AlphaFold algorithm developed by Google DeepMind.

Agnes Toth-Petroczy, who supervised the study, concludes, "We provide a basic framework for integrating clinical and protein data to aid assessing the impact of mutations. We hope that our tool and web server will ease variants effect assessment and clinical decision-making. Furthermore, the newly developed features can be applied to other genes and organisms beyond humans."