Fig. 6 |. Manual review misclassifications recovered by the deep learning model.

The Venn diagram illustrates variants identified as somatic by manual review (MR-specific), by both pipelines (Both), and by the deep learning classifier (Classifier-specific). For these three groups, the number of variants that have direct overlap with CIViC annotations and the total number of evidence items associated with all variants within each group are shown. These evidence items are parsed by those that convey variant sensitivity to a drug, variant resistance to a drug, variant that confers better or worse prognosis, variant that confers disease diagnosis, and variant that shows predisposing evidence for disease. The manual re-review panel shows the number of clinically relevant variants that agreed with the classifier call upon re-review by seven individuals.