Harnessing genotype and phenotype data for population-scale variant classification using large language models and bayesian inference DOI Creative Commons

Toby Manders,

Christopher Tan,

Yuya Kobayashi

и другие.

Human Genetics, Год журнала: 2025, Номер unknown

Опубликована: Апрель 23, 2025

Variants of Uncertain Significance (VUS) in genetic testing for hereditary diseases burden patients and clinicians, yet clinical data that could reduce VUS are underutilized due to a lack scalable strategies. We assessed whether machine learning approach using genotype phenotype improve variant classification VUS. In this cohort study multi-step approach, patient from test requisition forms were used distinguish with molecular diagnoses controls ("patient score"). A generative Bayesian model then scores classifications infer pathogenicity ("variant The included 3.5 million referred across various conditions. Primary outcomes model- gene-level discrimination, performance, probabilistic calibration, concordance orthogonal measures. Integration into semi-quantitative framework was based on posterior probabilities matching PPV ≥ 0.99/NPV 0.95 thresholds, followed by expert review. generated 1,334 models (CVMs); 595 showed high performance both steps (AUROCpatient 0.8 AUROCvariant 0.8) held-out data. High-confidence predictions these CVMs provided evidence 5,362 observed 200,174 patients, representing 23.4% all observations genes. 17 frequently tested genes, reclassified over 1,000 unique VUS, reducing report rates 9-49% per condition. conclusion, improved reduced

Язык: Английский

Experience using conventional compared to ancestry-based population descriptors in clinical genomics laboratories DOI Creative Commons
Kathryn E. Hatchell, Sarah Poll, Emily M. Russell

и другие.

The American Journal of Human Genetics, Год журнала: 2025, Номер unknown

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

1
The impact of systematized generation, evaluation, and incorporation of machine learning algorithms for clinical variant classification DOI Creative Commons
Laure Frésard, Flavia M. Facio, Elaine Chen

и другие.

medRxiv (Cold Spring Harbor Laboratory), Год журнала: 2025, Номер unknown

Опубликована: Фев. 6, 2025

Summary Variants of uncertain significance (VUS) pose a significant challenge for those undergoing genetic testing, leading to prolonged uncertainty and inappropriate medical care. VUS rate reduction is critical fully realize the utility testing all populations. With growth large-scale biological data sources modern Machine Learning (ML) techniques, predictive modeling has enormous potential reduction. For this purpose, we developed Invitae Evidence Modeling™ Platform (EMP), with key features designed maximize confidence algorithms variant classification. First, input new model curated correspond single major evidence category within classification framework. Second, gene-specific training and/or validation performed each type. Third, accuracy thresholds are set filter out models that do not meet stringent metrics. Finally, prediction scores pathogenicity calibrated ensure internally consistent weighting The EMP accelerated development ML greatly expanded amount available been applied more than 800,000 variants across 1 million individuals, 42% which would have without evidence. Importantly, definitive classifications (P, LP, LB, B) made high prospective concordance (>99%) ClinVar submissions. demonstrate further use reduce disparity race/ethnicity/ancestry (REA) groups.

Язык: Английский

Процитировано

0
Calibrated Functional Data Decreases Clinical Uncertainty for Tier 1 Monogenic Disease: Application to Long QT Syndrome DOI Creative Commons
Chai‐Ann Ng, Matthew J. O’Neill, Samskruthi Reddy Padigepati

и другие.

medRxiv (Cold Spring Harbor Laboratory), Год журнала: 2025, Номер unknown

Опубликована: Фев. 8, 2025

Abstract Rare missense variants are often classified as of uncertain significance (VUS) due to insufficient evidence for classification. These ambiguous findings create anxiety and frequently lead inappropriate workup, colloquially referred the ‘diagnostic odyssey’. Well-validated high-throughput experimental data have potential significantly reduce number VUS identified by clinical genetic testing, though extent this reduction optimal strategies achieve it remain unclear. 1

Язык: Английский

Процитировано

0
The Time Lag Associated With the Reclassification of Germline BRCA Variants' Pathogenicity Is Critical for Cancer Patients DOI Creative Commons

Tomomi Hayashi,

Hiroyuki Matsubayashi, Yoshimi Kiyozumi

и другие.

Cancer Science, Год журнала: 2025, Номер unknown

Опубликована: Март 4, 2025

The period for reclassifying the variants' pathogenicity is too long patients with advanced or recurrent cancers variants of uncertain significance that can be objectively re-evaluated to (likely) pathogenic according accumulated evidence. At present, we have three breast cancer germline variant BRCA2 c.7847C>T, which was genetically evaluated likely by paper in Cancer Science.

Язык: Английский

Процитировано

0
DTreePred: an online viewer based on machine learning for pathogenicity prediction of genomic variants DOI Creative Commons
Daniel H. F. Gomes, Inácio Gomes Medeiros, Tirzah Braz Petta Lajus

и другие.

BMC Bioinformatics, Год журнала: 2025, Номер 26(1)

Опубликована: Апрель 9, 2025

A significant challenge in precision medicine is confidently identifying mutations detected sequencing processes that play roles disease treatment or diagnosis. Furthermore, the lack of representativeness single nucleotide variants public databases and low rates underrepresented populations pose defies, with many pathogenic still awaiting discovery. Mutational pathogenicity predictors have gained relevance as supportive tools medical decision-making. However, disagreement among different regarding identification rooted, necessitating manual verification to confirm mutation effects accurately. This article presents a cross-platform mobile application, DTreePred, an online visualization tool for assessing variants. DTreePred utilizes machine learning-based model, including decision tree algorithm 15 learning classifiers alongside classical predictors. Connecting diverse prediction algorithms streamlines variant analysis, whereas enhances accuracy reliability data. integration information from various sources techniques aims serve functional guide decision-making clinical practice. In addition, we tested case study involving cohort Rio Grande do Norte, Brazil. By categorizing list oncogenes suppressor genes classified ClinVar inexact data, successfully revealed more than 95% integrity test 200 known yielded 97%, surpassing expected previous models. offers robust solution reducing uncertainty Improving assessments has potential significantly increase diagnoses treatments, particularly populations.

Язык: Английский

Процитировано

0
Harnessing genotype and phenotype data for population-scale variant classification using large language models and bayesian inference DOI Creative Commons

Toby Manders,

Christopher Tan,

Yuya Kobayashi

и другие.

Human Genetics, Год журнала: 2025, Номер unknown

Опубликована: Апрель 23, 2025

Variants of Uncertain Significance (VUS) in genetic testing for hereditary diseases burden patients and clinicians, yet clinical data that could reduce VUS are underutilized due to a lack scalable strategies. We assessed whether machine learning approach using genotype phenotype improve variant classification VUS. In this cohort study multi-step approach, patient from test requisition forms were used distinguish with molecular diagnoses controls ("patient score"). A generative Bayesian model then scores classifications infer pathogenicity ("variant The included 3.5 million referred across various conditions. Primary outcomes model- gene-level discrimination, performance, probabilistic calibration, concordance orthogonal measures. Integration into semi-quantitative framework was based on posterior probabilities matching PPV ≥ 0.99/NPV 0.95 thresholds, followed by expert review. generated 1,334 models (CVMs); 595 showed high performance both steps (AUROCpatient 0.8 AUROCvariant 0.8) held-out data. High-confidence predictions these CVMs provided evidence 5,362 observed 200,174 patients, representing 23.4% all observations genes. 17 frequently tested genes, reclassified over 1,000 unique VUS, reducing report rates 9-49% per condition. conclusion, improved reduced

Язык: Английский

Процитировано

0