Molecular evidence platform for auditable, continuous optimization of variant interpretation in genetic and genomic testing and analysis

Abstract

Disclosed herein are system, method, and computer program product embodiments for optimizing the determination of a phenotypic impact of a molecular variant identified in molecular tests, samples, or reports of subjects by way of regularly incorporating, updating, monitoring, validating, selecting, and auditing the best-performing evidence models for the interpretation of molecular variants across a plurality of evidence classes.

Description

BACKGROUND[0001]Molecular tests, such as genetic and genomic tests, are increasingly important diagnostic tools in a wide-array of clinical settings, from an individual's risk of neonatal seizures, abnormal heart rhythm (e.g., arrhythmia) or predisposition to developing cancers. The determination of the phenotypic impact (e.g., both clinical and non-clinical, including, but not limited to, pathogenicity, functionality, or relative effect) of a molecular variant—such as a genotypic (sequence) variant—identified within a subject is often the cornerstone of clinical molecular testing. The validity and utility of molecular testing can require that such determinations (e.g., often known as variant classifications) be evidence-based, objective, and systematic (Yandell et al. Genome Res. 2011 September; 21(9): 1529-42).[0002]Driven in large-part by next-generation sequencing (NGS) technologies, rapid advances in genetic and genomic technologies have led to dramatic increases in the volume ...

AI Technical Summary

Problems solved by technology

Even in the most heavily studied clinical genes and conditions, existing knowledge of the clinical significance of molecular variants often remains sparse.

For example, ˜50% of BRCA1 non-synonymous single-nucleotide genotypic (sequence) variants in a large public repository of clinical significance classifications (ClinVar) have conflicting classifications, and a consensus-based definition of truth can lead to a classification instability of ˜11% over a 12 month window (Landrum et al., 2015).

However, owing to the consistent growth and shifting nature of variant classifications—which form the basis of “truth sets” for the evaluation of evidence models—the computed performance metrics (e.g., diagnostic, classification, regression accuracy, etc.) for any evidence model are frequently outdated.

In addition, a reliance on a wide array of evidence models developed (e.g., computed, assayed, or aggregated) and evaluated in distinct settings (e.g., with frequently disjoint truth set definitions) often results in incoherent evaluation metrics among evidence models.

Together these factors complicate the evaluation and use of evidence models as support for variant interpretation.

As a consequence, a variant interpretation support system can not be able to reliably compare the performance of evidence models whereby evaluations are based on different data, within or between their different classes.

Thus, the variant interpretation support system can be unable to systematically and objectively compare the performance of the different evidence models.

While continued scientific work and publications routinely generate new evidence models, the lack of uniform “truth set” definitions, lack of synchronous updating, and biases in their performance evaluation (e.g., as might arise from authorship interests), can limit the inherent quality and utility of the evidence generated and their associated performance metrics.

As a consequence, a variant interpretation support system cannot reliably compare the performance of evidence models that were evaluated with different performance metrics, within or between their different classes.

In addition to these issues with evidence evaluation, the consistent growth and shifting nature of existing classifications (e.g., and hence truth sets) affects the robustness of evidence models, which often require a supervised learning step for generation.

As such, the variant interpretation support system can not have access to the most up-to-date evidence models possible.

Finally, the variant interpretation support system can be incapable of confirming that an evidence model was generated at a given moment in time, or generated with a given dataset.

A genetic and genomic test provider that obtains supporting evidence from the variant interpretation support system can therefore be unable to guarantee that performance metrics (e.g., diagnostic, classification, regression accuracy, etc.) for the evidence model are up-to-date, robust, and computed exclusively on disjoint data, e.g., on the basis of variants not used (or available) in the generation of the model.

Images