The ten year evolution of Sherloc, a points-based framework for genetic variant classification
Kobayashi, Y.; Westbrook, M.; Tan, C. A.; Marquez Luna, C.; Wahl, A.; Swartzlander, D.; Ouyang, K.; Daber, R. D.
Show abstract
ImportanceVariant classification underpins clinical genetics. We examine the impact of classification framework updates and the utility of semi-quantitative scores for interpreting variants of uncertain significance (VUS). ObjectiveEvaluate the impact of Sherloc on variant classification and VUS reduction in a large dataset. DesignCohort study using variant classification data generated between 2016 to 2025. Setting: A laboratory providing testing for hereditary genetic disorders. Participants5.5 million patients referred for genetic testing. Main Outcome(s) and Measure(s)We hypothesized that framework revisions would decrease the rate of VUS and improve accuracy. Outcome measures included changes in variant classification over time and in simulations. We further hypothesized that variant classification scores could predict the probability of pathogenicity and VUS reclassification rates. Outcome measures included the correlation of variant classification scores with the direction and rate of reclassification. ResultsSherloc has been used to classify 2.6 million variants across 19 successive versions (v4.2 to v7.1). Tracking 32,241 variants classified using v4.2, we found that 3,834 VUS had been reclassified by the end of the study period; 2801 (73%) of those relied on evidence criteria introduced after v4.2. In a simulation that removed various post-v4.2 evidence categories from 615,341 recent classifications, removing AI-related criteria alone changed 129,459 (21.04%) classifications. Classification reversals declined from 13/2909 (0.45%) in v4.2 to 7/61587 (0.011%) in v7.1, indicating improved accuracy concurrent with VUS reduction. The cumulative Sherloc score (pathogenic points minus benign points) strongly correlated with the log ratio of upgrades to downgrades (r{superscript 2} = 0.95), but less with the overall reclassification rate (r{superscript 2} = 0.49). Although VUS reclassified using RNA or cascade segregation analysis tended to have higher scores, the majority of reclassified variants had intermediate scores. Conclusion and RelevanceAdaptive, continuously updated classification systems that rapidly incorporate advances in clinical genetics, such as AI tools, reduce VUS and improve accuracy. Systematic performance monitoring, including reclassification tracking, enables data-driven improvements. A points-based framework like Sherloc represents progress toward probabilistic variant evaluation, which may lead to improved medical management.
Matching journals
The top 2 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Informing Variant Assessment using Structured Evidence from Prior Classifications (PS1, PM5, and PVS1 Sequence Variant Interpretation Criteria) 96%
- IGenomic answers for children: Dynamic analyses of >1000 pediatric rare disease genomes 96%
- A Framework for Automated Gene Selection in Genomic Screening 96%
Similar papers in this journal
- Extracting and calibrating evidence of variant pathogenicity from population biobank data 97%
- Advanced variant classification framework reduces the false positive rate of predicted loss of function (pLoF) variants in population sequencing data 96%
- Availability of benign missense variant “truthsets” for validation of functional assays: current status and a novel systematic approach 95%
Similar papers in this journal
- Evaluating Genome Sequencing Strategies: Trio, Singleton, and Standard Testing in Rare Disease Diagnosis 95%
- A systematic analysis of splicing variants identifies new diagnoses in the 100,000 Genomes Project. 95%
- From Text to Translation: Using Language Models to Prioritize Variants for Clinical Review 95%
Similar papers in this journal
- Long-read genome sequencing for the diagnosis of neurodevelopmental disorders 94%
- Leveraging TOPMed Imputation Server and Constructing a Cohort-Specific Imputation Reference Panel to Enhance Genotype Imputation among Cystic Fibrosis Patients 92%
- Omics-informed CNV calls reduce false positive rate and improve power for CNV-trait associations 92%
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.