P569: The diagnostic yield of exome and genome sequencing in a large and diverse cohort of over 300,000 patients: Context matters*

These proposed changes raise essential questions: Are computational algorithms sufficiently robust to justify upgraded weight?And, critically, could downgrading population-frequency evidence reduce specificity in distinguishing benign from pathogenic variants?Methods: We have extracted curated pathogenic/likely pathogenic (P/LP) and benign/likely benign (B/LB) variants from 4,000 clinical exomes processed in our laboratory.All variants were annotated in ClinVar with multiple submitters, no conflicts, and criteria provided.For each variant, we assessed whether the ACMG criteria PM2 (absence/rarity in population databases) or PP3 (computational evidence of deleterious effect) showed stronger discriminative performance.Comparative analyses evaluated the frequency, concordance, and classification accuracy of PM2 vs PP3 across variant classifications.Results: Among 1,009 P/LP nonsynonymous SNVs, most showed high deleteriousness scores, with 79% having REVEL > 0.644, 74% REVEL > 0.7, 29% had BayesDel >= 0.5, and 24% REVEL >= 0.932; only 6% scored REVEL < 0.4.P/LP variants were consistently rare: 959 (95%) had gnomAD4 AF < 0.001 and 340 (34%) AF < 0.00001.Among 15,244 B/LB nonsynonymous SNVs, 86% had REVEL < 0.4 and 76% BayesDel < -0.16, while 4% exceeded REVEL > 0.644 and 36 (0.2%) B/LB variants had a BayesDel >= 0.5.4,659 (30.6%)B/LB variants met AF < 0.001, and only 63 (0.4%) met AF < 0.00001.Conclusion: Ongoing SVC_v4 discussions propose reducing the strength of population-based evidence while increasing the weight of computational prediction criteria.Our findings indicate that population frequency remains one of the most powerful discriminators in variant interpretation.Although PP3 metrics such as REVEL and BayesDel showed reasonable performance, population thresholds (particularly ultra-rare cutoffs, now more accessible through expanded population databases) demonstrated markedly clearer separation between pathogenic and benign variants.These results highlight the continued reliability and clinical value of population data.Overall, our analysis supports maintaining PM2 as a meaningful evidentiary criterion rather than reducing its strength in favor of expanded reliance on computational prediction tools.