Improving decision support for uncertain gene varients

Rapidly evolving technologies such as chip arrays and next-generation sequencing are uncovering human genetic variants at an unprecedented pace. Unfortunately, this ever growing collection of gene sequence variation has limited clinical utility without clear association to disease outcomes. As electronic medical records begin to incorporate genetic information, gene variant classification and accurate interpretation of gene test results plays a critical role in customizing patient therapy. To verify the functional impact of a given gene variant, laboratories rely on confirming evidence such as previous literature reports, patient history and disease segregation in a family. By definition variants of uncertain significance (VUS) lack this supporting evidence and in such cases, computational tools are often used to evaluate the predicted functional impact of a gene mutation. This study evaluates leveraging high quality genotype-phenotype disease variant data from 20 genes and 3986 variants, to develop gene-specific predictors utilizing a combination of changes in primary amino acid sequence, amino acid properties as descriptors of mutation severity and Naïve Bayes classification. A Primary Sequence Amino Acid Properties (PSAAP) prediction algorithm was then combined with well established predictors in a weighted Consensus sum in context of gene-specific reference intervals for known phenotypes. PSAAP and Consensus were also used to evaluate known variants of uncertain significance in the RET proto-oncogene as a model gene. The PSAAP algorithm was successfully extended to many genes and diseases. Gene-specific algorithms typically outperform generalized prediction tools. Characteristic mutation properties of a given gene and disease may be lost when diluted into genomewide data sets. A reliable computational phenotype classification framework with quantitative metrics and disease specific reference ranges allows objective evaluation of novel or uncertain gene variants and augments decision making when confirming clinical information is limited.