R43HG012535

PROJECT SUMMARY Entire patient genomes can now be sequenced for hundreds of dollars, and the price is still falling. Physicians now routinely order large gene panels as a way of diagnosing disease and guiding treatment. While the widespread use of these tests is beneficial for patient care, it also introduces a challenge of large-scale data interpretation. Currently, most unique variants uncovered by genetic tests have insufficient evidence for confident classification. These “variants of unknown significance” (VUS) hinder timely diagnosis and treatment of deadly diseases such as heart disease and cancer. An improved and proactive approach is needed to decrease the number of variants that are classified as VUS. Functional assays performed in laboratories are important sources of evidence used for classification of gene variants. Historically, these experiments have been low-throughput, generating data for one variant at a time. Furthermore, such experiments are reactive, meaning they are performed only after a given variant has been observed in the clinic. To proactively expand genetic variant characterization, several academic laboratories have recently developed Multiplexed Assays of Variant Effect (MAVEs), which collect data on thousands of protein variants in a single experiment. MAVEs hold great promise as a source of high-throughput functional evidence. Nonetheless, there are currently no commercial platforms that curate and robustly analyze the large and growing number of MAVE datasets being generated by academic labs to inform clinical variant interpretations. As a result, the potential for these data to inform lifesaving medical decisions is unrealized. To address the need for improved clinical variant interpretation, Constantiam Biosciences is developing VarifyTM, a first of its kind platform specializing in the translation of MAVE data into actionable information to support clinical variant interpretation. Varify brings two key innovations to the field of genomic interpretation: the application of Bayesian inference, which is the best proven method for handling uncertainty, and probabilistic programming, a novel computational technique that allows statistical inference to be performed efficiently on models that accurately reflect the conditions under which the data were generated. To support the Phase I program, Constantiam Biosciences has developed an early-stage prototype of Varify. The company will build upon these preliminary efforts to execute the Phase I SBIR program with the goal of developing and assessing Varify’s variant effect inference framework. Aim 1 is focused on augmenting the existing early-stage variant effect inference framework to include modules that model the influence of signal-corrupting processes present in MAVE experiments that can distort and obscure variant effects. The expanded framework will be continuously evaluated using simulated data (Aim 1) and applied on existing MAVE data sets for BRCA1 and PTEN (Aim 2). Successful completion of these aims will provide critical proof-of-concept for Varify’s expanded framework and support a Phase II program that will apply Varify more broadly and develop a commercial-ready product.