Predicting the Pathogenicity of Variants of Uncertain Significance in BRCA1 and BRCA2 by Cross-Species Evolutionary Patterns Using Machine Learning

A major challenge in clinical diagnostics is estimating the pathogenicity of detected variants of uncertain significance (VUSs). This challenge entails major clinical implications, for example the pathogenicity of VUSs in BRCA1 directly translate into the risk to develop hereditary breast and ovarian cancer, which for high risk cases requires preemptive-surgical or chemotherapeutic treatments. Therefore, accurate estimation of the pathogenicity of VUSs is necessary improve patient outcomes. One successful approach to predicting VUS pathogenicity is to estimate the functionality of variants based on their conservation-rate across species. Here we present a novel approach for the assessment of VUSs by analyzing nucleotide evolutionary patterns (EvoPatterns). We hypothesized that additional knowledge can be extracted, not only by inspecting the rate of conservation, but also by identifying informative conservation patterns of nucleotides across species. To generate EvoPatterns, we identified a variety of evolutionary patterns by profiling the conservation patterns of BRCA1 nucleotides across 100 vertebrates. We then used machine-learning to uncover relationships between these evolutionary patterns and the pathogenicity of known variants in BRCA1. EvoPatterns is the first to incorporate complex evolutionary patterns into prediction tools. We show that EvoPatterns successfully captures the complex nature of the evolution of nucleotides in BRCA1. Moreover, it correctly predicts the clinical implications of known BRCA1 mutations better than established conservation methods. As the genomic data of patients, disease models and species continues to grow exponentially, we expect that EvoPatterns will further improve in prediction accuracy and enable better association of evolutionary patterns and phenotypes.