Benchmark analysis
Adopting ACMG/AMP guidelines for genetic hearing loss, 152 ACMG/AMP rules were activated for 50 pilot variants curated by the ClinGen HL-EP. Of these, 55% (83/152) rules could be automatically interpreted by VIP-HL, spanning PVS1, PM1, PM2, PM4, PM5, PP3, BA1, BS1, BP4, and BP7. Overall, VIP-HL achieved 96% (80/83) concordant interpretations compared with activations by ClinGen HL-EP (Figure 1a). Of 83 activated rules (Figure 1b), the three discordant activations between VIP-HL and ClinGen HL-EP were PM2_Supporting (2 times) and BA1 (1 time). Those discrepancies were most likely due to using an updated version of gnomAD and the adoption of popmax filtering allele frequency (Whiffin et al. 2017) by VIP-HL (Table 1, variant #1-3). By comparison, InterVar only achieved 47% (39/83) concordant activations, significantly lower than that by VIP-HL (Figure 1b).
Both VIP-HL and InterVar activated certain rules that ClinGen HL-EP did not apply. Specifically, VIP-HL activated only four rules six times, including PM1 (2 times), PP3 (2 times), PM2 (1 time), PVS1 (1 time), BS2 (2 times) (Table 1, variant #4-8). InterVar, however, activated 11 rules 100 times (Supplementary Figure S1), spanning PVS1 (2 times), PM1 (15 times), PM2 (20 times), PP2 (2 times), PP3 (12 times), PP5 (25 times), BS1 (2 times), BS2 (8 times), BP1 (11 times), BP4 (1 times), BP6 (2 times). Notably, four rules (PP2, PP5, BP1, and BP6) that were considered not applicable for genetic hearing loss accounted for 40% of the activations (40/100). These results demonstrated the reliable annotations by VIP-HL and reinforced the importance of disease-specific annotations in variant interpretation.