Molecular evidence platform for auditable, continuous optimization of variant interpretation in genetic and genomic testing and analysis
Inventors
Colavin, Alexandre • Araya, Carlos L. • Reuter, Jason A.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12136472-B2
Publication Date
2024-11-05
Expiration Date
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.
Core Innovation
The invention is a molecular evidence platform that continuously optimizes variant interpretation by regularly incorporating, updating, validating, selecting, and auditing best-performing evidence models across multiple evidence classes. The platform records and uses evidence models comprising evidence data that yield predictions of phenotypic impacts of molecular variants for a target entity that includes a functional element.
Validation performance data are evaluated based on production data derived from clinical data and/or population data using a uniform set of performance metrics to provide an unbiased estimate of predictive performance at a given time. Test performance data are evaluated based on evidence data and test data received for the evidence model, where the test data includes molecular variants disjoint from those in the production data and having phenotypic impacts that are unknown or unavailable at the time of evidence model generation.
The test performance data are computed with the same uniform set of performance metrics to provide an unbiased estimate of predictive performance at a later time, and the best-performing evidence model is selected via ranking to provide predicted phenotypic impact in response to a query. A key feature is auditable provenance for supporting data, where supporting data for the evidence model are hashed and associated with audit records stored in a distributed structure such as a blockchain, Merkle tree, or time-stamped public ledger.
Later verification of data availability, content, and creation time is enabled, including certificates of validation for audit records, with cryptographic proof via Merkle tree roots and proofs.
Claims Coverage
The independent claims cover a computer-implemented method, a non-transitory computer-readable medium, and a system for predicting phenotypic impact of a molecular variant of interest. Across the independent claims, the workflow includes recording an evidence model, evaluating validation performance on production data with a uniform set of performance metrics for unbiased performance at a given time, evaluating test performance on disjoint test data with the same uniform metrics for unbiased performance at a later time, ranking evidence models, and providing predicted phenotypic impact via a variant interpretation terminal.
Evidence model recording for phenotypic impact prediction
Recording an evidence model comprising evidence data, wherein the evidence data comprises objects, algorithms, and/or functions that yield predictions of phenotypic impacts of molecular variants for a target entity, and wherein the target entity comprises a functional element.
Unbiased validation performance from production data with uniform metrics
Evaluating validation performance data for the evidence model based on production data, wherein the production data represents a set of molecular variants with associated phenotypic impacts derived from clinical data and/or population data, and wherein the validation performance data corresponds to a uniform set of performance metrics computed using the production data and evaluating validation performance data gives an unbiased estimate of the predictive performance of the evidence model at a given time.
Unbiased test performance from disjoint test data with uniform metrics
Evaluating test performance data for the evidence model based on the evidence data and test data in response to receiving the test data for the evidence model, wherein the test data comprises a set of molecular variants with associated phenotypic impacts derived from clinical data and/or population data, wherein said set of molecular variants are disjoint from those in the production data and includes phenotypic impacts for molecular variants of unknown phenotypic impacts at the time of evidence model generation, or unavailable at the time of evidence model generation, and wherein the test performance data corresponds to the uniform set of performance metrics computed using the test data and evaluating the test performance data gives an unbiased estimate of the predictive performance of the evidence model at a later time.
Ranking evidence models and providing predicted phenotypic impact via terminal
Ranking the evidence model in a set of evidence models for the target entity based on the validation performance data and the test performance data; and providing the predicted phenotypic impact using a best-performing evidence model for the target entity based on the ranking in response to a query for the predicted phenotypic impact of the molecular-variant of interest for the target entity from a variant interpretation terminal.
Hashed supporting data stored with associated hash-to-supporting-data mapping
Generating a hash value of supporting data for the evidence model, wherein the supporting data comprises the evidence data, the validation performance data, and/or the production data; wherein the hash value of the supporting data for the evidence model is stored in a database, wherein the database associates the hash value with the supporting data.
Across the independent claims, the core claim coverage centers on producing unbiased estimates of predictive performance by computing a uniform set of performance metrics on both production-derived validation data (for a given time) and disjoint test data with unknown or unavailable phenotypic impacts at generation time (for a later time), then ranking evidence models and outputting predicted phenotypic impact via a variant interpretation terminal. The claims further include recording evidence models that produce phenotypic impact predictions and, in one independent claim, hashing supporting data and storing hash values with associated supporting data in a database.
Stated Advantages
Gives an unbiased estimate of the predictive performance of the evidence model at a given time based on production data using uniform performance metrics.
Gives an unbiased estimate of the predictive performance of the evidence model at a later time based on disjoint test data using the uniform set of performance metrics.
Provides predicted phenotypic impact using a best-performing evidence model selected based on validation and test performance.
Documented Applications
No documented applications found
Interested in licensing this patent?