A variant by any name: quantifying annotation discordance across tools and clinical databases

Area of Science:

• Genomic Medicine
• Bioinformatics
• Clinical Diagnostics

Background:

• Clinical genomic testing relies on accurate variant identification and reporting for disease association.
• High-throughput sequencing generates vast amounts of data, posing challenges for cross-referencing variants.
• Discrepancies exist between genomic position-based variant calls and transcript/protein-based resource descriptions.

Purpose of the Study:

• To evaluate the accuracy of SnpEff, Variant Effect Predictor, and Variation Reporter in generating variant nomenclature.
• To assess the concordance of variant annotations between these tools and major databases (ClinVar, COSMIC).
• To identify inconsistencies in variant representation for clinical genomic interpretation.

Main Methods:

• Evaluated three variant annotation tools against a manually curated set of 126 variants with HGVS-compliant descriptors.
• Compared tool-generated transcript and protein variant nomenclature derived from genomic coordinates.
• Assessed concordance between SnpEff/Variant Effect Predictor annotations and ClinVar (germline) and COSMIC (cancer) databases.

Main Results:

• Substantial discordance observed in describing insertions/deletions between annotation tools and databases.
• Accuracy for coding and protein changes ranged from 50-90%, with lower accuracy for non-SNV variants.
• Exact concordance for Single Nucleotide Variants (SNVs) was high (>99.5%) with ClinVar but lower (<88%) with COSMIC, especially for insertions (<15%).

Conclusions:

• Significant inconsistencies in variant representation across tools and databases hinder variant matching and classification.
• Syntax differences, while potentially clear to clinicians, can confound automated variant interpretation.
• Urgent need for uniform standards in variant annotation, including consistent genomic reference reporting, for reliable clinical care.