Researchers from Myriad Genetics have published a paper in Clinical Genetics that describes the informatics components of its MyVision variant classification program, which the company uses to categorize mutations of unknown significance as either pathogenic or benign.
Myriad’s pipeline uses data from multiple sources, assays, and various statistical tools to categorize unknown variants identified in genetic tests as either pathogenic or benign and to reclassify mutations as new evidence about them becomes available. Although its paper focuses specifically on an application of the pipeline to cancer genes — specifically BRCA 1/2 genes which have mutations associated with hereditary breast and ovarian cancers — similar techniques could be used, with a few alterations, to classify variants of unknown significance (VUS) in other genes, the researchers wrote.
In a statement, Richard Wenstrup, chief medical officer at Myriad Genetics Laboratories and a co-author on the paper, claimed that his company's methods have "significantly reduced" the rates of unknown variants for patients across ethnic groups. For example, the company reports that among patients of African ancestry, it has reduced the VUS rate from 38.6 percent to 3.6 percent and for patients with Latin ancestry, the VUS rate has dropped from 26.1 percent to 3 percent. Furthermore, the researchers wrote that Myriad's methods have "lowered the percentage of [its] tests in which one or more BRCA1/2 variants of uncertain significance are detected to 2.1 [percent]" — that’s down from roughly 13 percent of all BRCA 1/2 tests a decade ago.
Myriad's protocol for classifying new variants, according to the paper, includes a classification step based on guidelines from the American College of Medical Genetics as well as internal guidelines set by the company's New Mutation Committee — a group that includes laboratory directors, clinical variant specialists, genetic counselors, and experts in statistical genetics, biochemistry, and structural biology. The company also taps into public databases and scientific literature to keep updated on new information about variants of interest as it becomes available and to reclassify them if necessary.
Statistical techniques in the pipeline include the Mutation Co-occurrence (MCo) algorithm, one of two proprietary classification tools developed by Myriad's bioinformatics and variant classification teams. The Clinical Genetics paper explains that the MCo takes advantage of observations from previous studies that if a known pathogenic mutation shows up in familial genomes, it usually is the "primary cause" of whatever disease is present and its presence "reduces the likelihood that a VUS in the same pathway is clinically relevant."
The second proprietary tool in the pipeline is called Pheno, a history-weighting algorithm that compares the family histories of patients with VUS to that of families with known harmful variants and those with no mutations, and is based on the premise "that individuals with deleterious mutation are expected to have more severe personal and family histories than individuals with benign polymorphisms," the authors wrote.
Ronald Rogers, Myriad's executive vice president for corporate communications, explained to BioInform in an email that both MCo and Pheno rely on data that Myriad obtained from clinically testing more than 400,000 patients.
"Pheno uses this dataset to determine which variants are associated with more severe personal and family histories of cancer, consistent with the variant being a deleterious mutation, and which variants do not appear to be associated with increased familial cancer risk," he said. MCo, meanwhile "statistically measures the significance of a variant being identified in one or more individuals who are also known to carry a deleterious mutation in addition to the variant in question. The more often a particular variant is shown to co-occur with a deleterious mutation, the less likely that variant is to be deleterious itself."
Other lines of evidence the company uses to assess variants' pathogenicity include looking at the frequency with which they appear in populations, with variants being classified as benign if they appear in more than two percent of a control population of more than 200 individuals, according to the paper. As part of this step, "we also statistically compare the frequency of a specific variant in the general population with the frequency of the same variant within a high-risk patient population" since "overrepresentation of a variant within a high-risk patient population may indicate that the variant is associated with increased cancer risk," Rogers said.
The protocol also includes a segregation analysis step, where researchers look for evidence that either supports or contradicts a particular variant's association with cancer in one or more families. Myriad uses a "modified" version of traditional segregation analysis — which requires testing multiple members of large families of a size not often seen in a clinical setting — that "allows us to combine data from multiple smaller families in order to reclassify variants," Rogers said.
Also used to a limited degree is evolutionary conservation analysis where researchers compare protein sequences across species. Myriad notes on its website that because of high false-positive rates, it only uses this sort of analysis to downgrade variants formerly listed as pathogenic but does not classify variants as harmful or even potentially harmful on the basis of conservation evidence alone.
In response to a question about how Myriad's methods compare to other approaches, Rogers said that while MCo and Phen have been "extensively" tested and validated internally to "ensure positive and negative predictive values of greater than 0.99" most publically available analysis tools have not had the same treatment.
"Most publicly available classification tools were developed for research, not clinical use, and their performance characteristics, such as positive and negative predictive values, are not known," he said. "We have performed limited analyses of some of these public tools and have found them to be highly inaccurate and inappropriate for clinical use."
Furthermore, "it is important to note that these performance characteristics are likely to be highly gene-dependent and it would be necessary to validate each tool on a gene-by-gene basis in order to determine the accuracy of these tools for a particular gene," he added.