New functional modeling platform reduces uncertainty in genetic testing

Genetic testing through DNA sequencing can detect millions of places where one person’s genome differs from another’s. Most of the time, these differences are harmless and deemed benign. Occasionally, they are the cause of disease or a marker of increased risk for a disease and deemed pathogenic. Differentiating between the benign and the pathogenic is a major purpose—and challenge—of medical genetics. Around half of individuals who get their DNA tested for diagnostic purposes are told they possess one or more DNA variant(s) of uncertain significance (VUS) because there is not enough evidence to determine whether these variants are likely to cause disease. To reduce the number of VUS, Invitae has added a new capability within its Sherloc variant interpretation framework. This advance in variant interpretation—our functional modeling platform (FMP)—is a novel system unique to Invitae that can accurately predict for many variants how they will affect gene function by pulling in many lines of evidence from both lab experiments and computational analyses. FMP changes genetic testing outcomes for 1 in 40 (or 2.5%) of all patients tested at Invitae. For these patients and their providers, this means that they will get a definitive answer, instead of a VUS. 

From its outset, the field of medical genetics has gathered evidence from many clinical and scientific sources to determine whether a given DNA variant is pathogenic or benign. At Invitae, we gather evidence from the rich population data provided by large public genome sequence repositories, from computational predictions of how a variant will affect the resulting protein’s function, from clinical records, and from peer-reviewed scientific studies. These methods have enabled Invitae to provide genomic information to more than 1 million individuals who need that information to guide their healthcare. As Invitae moves toward assisting millions more patients in the years ahead, we need additional sources of evidence that can lead to definitive results for more people.

FMP enhances our ability to interpret DNA variants by integrating several lines of evidence from disparate sources. FMP learns from a suite of discovery tools, which run the gamut from in vitro experiments in the lab to computational analysis of how proteins have evolved over time. This technology platform integrates biochemical, cellular, and computational data and feeds them to machine learning software that continuously updates and reanalyzes its predictions based on real-world outcomes and in silico experiments. The result is a significant enhancement of our ability to interpret DNA variants. 

FMP includes several individual components that independently generate evidence to inform variant interpretation. One is a highly efficient and coordinated approach to experiments called deep mutational scanning. First, we create mutations one-by-one at each DNA letter along the entire sequence of a gene. Then we monitor the effects of each mutation in thousands of simultaneous, small-scale experiments. This allows us to pinpoint the DNA variants most sensitive to disruption and determine whether they disrupt important biological processes. FMP also draws from existing knowledge of the physical properties of proteins to predict whether a DNA variant observed in a patient will be harmless or have detrimental effects. FMP even draws from evolutionary biology to understand which DNA sequences are preserved over many species and which are evolutionary dead ends. This knowledge can help FMP predict which variants are likely to disrupt protein function and be pathogenic and which are likely tolerated and benign. Moreover, the computational predictions are customized for each gene, making them more accurate than predictions from standard methods that typically apply the same rules to all genes. 

The platform has built-in quality control. The performance of each line of evidence is continuously tracked for its accuracy in variant interpretation. When relevant FMP predictions are integrated alongside other lines of evidence (e.g., clinical data), our Sherloc variant interpretation framework can achieve >99% accuracy for determining whether a variant is pathogenic or benign. This level of accuracy exceeds guidelines from the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. 

As of spring 2020, we have incorporated FMP evidence on ~2,500 VUS across all clinical areas, enabling us to reclassify VUS results for more than 5,000 patients who now have definitive answers from their genetic tests instead of uncertainty. Over time, we anticipate that the scope and performance of FMP will expand, as the system can readily accept new sources of evidence that give insight into the impact of variants on biological function and human health. 

To learn more about how Invitae delivers deeper, more informative results with FMP, read our recent white paper