Biomedical Informatics and Machine Learning for Clinical Genomics.

Author information

Abstract

While tens of thousands of pathogenic variants are used in the myriad clinical applications of genomics, there remains limited quantitative risk information (i.e. penetrance) or consensus about the clinical utility of the majority of extant variants and tests. Relatedly, with the rising demand for genetic counseling, there is a need for computational approaches that scale the ability of counselors to interpret new and existing variants, whether by automating the determination of pathogenicity or filtering variants too common in the general population to be sufficiently penetrant. To address these challenges, researchers are increasingly turning to integrative informatics approaches that leverage vast sources of data including electronic health records (EHRs) and population-level allele-frequency databases (e.g. gnomAD) as well as statistical and machine learning methods (e.g. support vector regression, deep learning). In this review, we highlight recent examples of these informatics and machine learning approaches that are improving our understanding of pathogenic variation, and discuss obstacles such as bias and interpretability that likely must be overcome for such approaches to adopt a more central role in clinical genomics.