Variant Classification Discordance: Contributing Factors and Predictive Models.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• The ClinVar database aggregates human variant data, combining submissions into multisubmitter records.
• Variant classification discordance between submitters results in a 'conflicting' label within ClinVar.

Purpose of the Study:

• To identify characteristics associated with conflicting variant classifications in ClinVar.
• To develop a predictive model for classification discordance in single-submission variants.

Main Methods:

• Utilized ClinVar data to analyze factors linked to classification conflicts.
• Employed the Extreme Gradient Boosting algorithm to train a classifier model.
• Evaluated model performance using accuracy, precision, recall, and F1-score on a test set.

Main Results:

• Population allele frequency, gene, variant type, protein consequence, deleteriousness score, first submitter, and submission count were associated with classification conflict.
• The optimized classifier achieved 88% accuracy, with weighted averages of 0.84 (precision), 0.88 (recall), and 0.85 (F1-score).
• Allele frequency, gene type, and the first submitter's identity showed strong associations with variant classification discordance.

Conclusions:

• The study successfully predicted discordance status for single-submitter variants in ClinVar.
• This predictive approach can help assess whether new single-submission variants align with or conflict with existing entries.
• Findings can assist clinical laboratories in variant assessment and interpretation.