Reassessment of pathogenic variants in open genetic databases highlights caveats in interpretation of actionable mutations

Introduction: The rising usage of next generation sequencing for diagnostic purposes necessitates routine interpretation of genomic data by clinicians. Open genetic databases are founded on decades of research, allocating more than 300,000 variants into clinically relevant categories.

Methods: We compare data in ‘Clinvar’ and ‘OMIM’ registries to frequency data in the genome aggregation database ‘gnomAD’.

Results: Comparison of allele frequency (AF) with disease prevalence shows that many variants presently defined as pathogenic are either mislabeled, induce non-disease traits or are not causative and should be classified as benign, in accordance with ACMG guidelines. Parsing the data to separate populations further enables detection of variants that are common in specific cohorts, yet not widespread in the general population, and are therefore mis-interpreted as pathogenic. We reclassify 106 “pathogenic” variants with AF>0.01, of which 58 are non-pathogenic (mean ratio of AF to disease prevalence 4466.96) and 22 cause a benign trait. Of the 106 variants, 69 were actionable and 16 might be considered for termination of pregnancy in prenatal screening. We provide examples of misclassification of severe diseases in patients studied for unrelated monogenic traits, demonstrate that pathogenicity prediction tools show low ability to evaluate population-level data, and show that functional molecular evidence of pathogenicity might be misleading.

Conclusions: Non-discriminatory use of open-source variant databases for clinical practice could potentially hold risk. Furthermore, AF data, including that in specific cohorts evaluated, should be routinely taken into consideration in determining pathogenicity of genomic variants, even if labeled as pathogenic in existing databases.