As the compendium of putatively disease causing variants expands, gathering the most current and accurate information is critical to computing variant classifications. Manual curation is laborious, time consuming and error prone, as information critical to variant interpretation may be missed. The QIAGEN Knowledge Base includes the most extensively curated database of variant specific publications, as well as data from other sources such as ClinVar, HGMD, CentoMD and OMIM. This knowledge base functions as the cornerstone of our clinical decision support tool, QIAGEN Clinical Insight (QCI), facilitating rapid variant filtering and prioritization, automated ACMG classification and reporting. Here, we compare the concordance of QCI’s automated variant classification with the ENIGMA expert panel assessments of BRCA1 and BRCA2 variants. As of December 2017, ClinVar contained 6154 expert reviewed BRCA1 and BRCA2 variants. These variants were exported and the resulting VCF was uploaded to QCI. QIAGEN curated content within the tool was used to automatically compute ACMG variant classifications and provide underlying evidence. QCI computed classifications were compared with ENIGMA assessments and demonstrated an extremely high rate of concordance, with 81.2% of classifications perfectly matching ENIGMA’s assessments. Moreover, the majority of differences observed would not affect clinical management, occurring either within Pathogenic/Likely Pathogenic or Benign/Likely Benign (11.1%) or Benign/Likely Benign vs VUS (7.2%), bringing the degree of concordance at the level of clinical actionability to 99.6%. Only 3 variants were significantly discordant (Benign/Likely Benign by ENIGMA vs. Likely Pathogenic by QCI). These included synonymous variants with functional data supporting splicing defects not considered by ENIGMA. In summary, with respect to clinical actionability, QCI automated ACMG classification was 99.6% concordant with ENIGMA expert panel variant assessments. This level of accuracy speaks to the quality of the clinical, functional, and population level data in the knowledge base, as well as the robustness of the underlying algorithm used to apply the ACMG guidelines.
J.L. Poitras, D. Richards, H. Su, T. Love, R. Yip