Population frequency modeling for quantitative variant pathogenicity estimation
Inventors
Manders, Toby • Nykamp, Keith • Colavin, Alexandre • Kobayashi, Yuya
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12191001-B2
Publication Date
2025-01-07
Expiration Date
Abstract
Embodiments of the disclosed technologies apply a logistic regression model to a set of population data for a set of genes. The set of population data includes a set of features for a variant located at a position within a gene. The set of features includes at least one population frequency meta-feature. The at least one population frequency meta-feature quantifies predictive value of allele frequency in the gene. Using the first set of population data, a variant classification prediction output by the logistic regression model is evaluated based on an expected variant classification. The logistic regression model is adjusted until at least one first performance criterion is satisfied to produce a trained logistic regression model. The trained logistic regression model is capable of outputting variant pathogenicity estimates that satisfy at least one second performance criterion.
Core Innovation
The disclosed invention configures a machine learning model by applying a logistic regression model to population data for genes in order to classify genetic variants and estimate variant pathogenicity. For each variant located within a gene, the method uses a set of features that includes at least one gene-level feature, at least one variant-level feature, and at least one population frequency meta-feature, along with a reference label indicating whether the variant is benign or pathogenic. The population frequency meta-feature quantifies predictive value of allele frequency in the gene.
The population frequency meta-feature is constructed to mathematically combine gene-level constraint and allele frequency. The method includes computing a gene-level constraint, including the gene-level constraint in the at least one gene-level feature, computing an allele frequency, including the allele frequency in the at least one variant-level feature, and including, in the at least one population frequency meta-feature, a mathematical combination of the gene-level constraint and the allele frequency. The logistic regression model is then applied to the set of features including that mathematical combination to output variant pathogenicity estimates.
The model is trained by evaluating variant classification predictions against expected variant classification indicated by the reference label and iteratively adjusting parameters or coefficients until output of a loss function satisfies at least one first performance criterion. The trained logistic regression model is capable of outputting variant pathogenicity estimates that satisfy at least one second performance criterion. Additional disclosed feature constructions include computing an exponent of a ratio of synonymous missense variants to nonsynonymous missense variants for inclusion in the population frequency meta-feature, and computing a mathematical combination of subpopulation frequency data and population frequency data for inclusion in variant-level features.
Claims Coverage
The partial content includes six independent claims: clm-00001, clm-00007, clm-00010, clm-00017 (system), clm-00023 (system), clm-00026 (system), plus machine-readable medium claims clm-00033, clm-00039, and clm-00042. Across these, the claims cover configuring and using a logistic regression model with gene-level features, variant-level features, and population frequency meta-features, including engineered mathematical combinations that relate gene-level constraint to allele frequency and, in specific claims, additional exponents and subpopulation/population combinations.
Logistic regression with gene-level constraint and mathematical combination meta-feature
Applying a logistic regression model to a first set of population data for a first set of genes using features comprising at least one gene-level feature, at least one variant-level feature, and at least one population frequency meta-feature, with a reference label indicating whether the variant is benign or pathogenic; computing a gene-level constraint; including the gene-level constraint in the at least one gene-level feature; computing an allele frequency; including the allele frequency in the at least one variant-level feature; including, in the at least one population frequency meta-feature, a mathematical combination of the gene-level constraint and the allele frequency; and applying the logistic regression model to the set of features including the mathematical combination to output variant pathogenicity estimates that satisfy at least one second performance criterion.
Logistic regression training and evaluation against reference labels with loss-based criteria
For each item of the first set of population data, evaluating a variant classification prediction output by the logistic regression model based on an expected variant classification indicated by the reference label; and iteratively adjusting a value of at least one parameter or coefficient of the logistic regression model until output of a loss function computed based on the variant classification prediction output by the logistic regression model satisfies at least one first performance criterion, producing a trained logistic regression model capable of outputting variant pathogenicity estimates that satisfy at least one second performance criterion.
Exponent of synonymous-to-nonsynonymous missense ratio in population frequency meta-feature
Computing, for the gene, an exponent of a ratio of synonymous missense variants to nonsynonymous missense variants; including the exponent of the ratio of synonymous missense variants to nonsynonymous missense variants in the at least one population frequency meta-feature; and applying the logistic regression model to the set of features including the exponent of the ratio of synonymous missense variants to nonsynonymous missense variants to output variant pathogenicity estimates that satisfy at least one second performance criterion.
Subpopulation and population frequency mathematical combination as a variant-level feature
Computing, for a variant, a mathematical combination of subpopulation frequency data and population frequency data; including the mathematical combination of subpopulation frequency data and population frequency data in the at least one variant-level feature; and applying the logistic regression model to the set of features including the mathematical combination of subpopulation frequency data and population frequency data to output variant pathogenicity estimates that satisfy at least one second performance criterion.
System implementing the population-frequency logistic regression approach
A system comprising at least one processor and at least one memory with instructions causing the processor to apply a logistic regression model to a first set of population data for a first set of genes using features comprising at least one gene-level feature, at least one variant-level feature, and at least one population frequency meta-feature with a reference label indicating whether the variant is benign or pathogenic; including computing a gene-level constraint, including it in the gene-level feature, computing an allele frequency and including it in the variant-level feature, and including a mathematical combination of the gene-level constraint and the allele frequency in the population frequency meta-feature; evaluating predictions against expected variant classification indicated by the reference label; and adjusting parameters or coefficients until at least one first performance criterion is satisfied to produce a trained logistic regression model capable of outputting variant pathogenicity estimates satisfying at least one second performance criterion.
Non-transitory machine-readable medium executing the population-frequency logistic regression approach
At least one non-transitory machine-readable medium comprising at least one instruction executable by at least one processor to perform operations including applying a logistic regression model to a first set of population data for a first set of genes using features comprising at least one population frequency meta-feature and a reference label indicating whether the variant is benign or pathogenic, wherein the population frequency meta-feature quantifies predictive value of allele frequency in the gene; computing a gene-level constraint; including the gene-level constraint in the at least one gene-level feature; computing an allele frequency; including the allele frequency in the at least one variant-level feature; including a mathematical combination of the gene-level constraint and the allele frequency in the at least one population frequency meta-feature; applying the logistic regression model to output variant pathogenicity estimates satisfying at least one second performance criterion; evaluating predictions; and adjusting parameters or coefficients until at least one first performance criterion is satisfied to produce the trained logistic regression model.
Across the independent claims, coverage centers on configuring and training a logistic regression model for variant classification/pathogenicity using gene-level features, variant-level features, and population frequency meta-features that quantify predictive value of allele frequency. The strongest differentiators are engineered population frequency meta-features derived from a mathematical combination of gene-level constraint and allele frequency, optional inclusion of an exponent of a ratio of synonymous missense variants to nonsynonymous missense variants, and optional inclusion of a mathematical combination of subpopulation frequency data with population frequency data as a variant-level feature. The claims further require iterative adjustment of logistic regression parameters based on a loss function and performance criteria.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?