Comparative evaluation of manual and automated ACMG/AMP variant classification: implications for clinical genetic practice.

Criterion-level concordance benchmark between tools (Franklin, VarSome, MobiDetails, GeneBe, InterVar, VarChat) and dual independent expert curation.

This study benchmarks several automated and AI-assisted tools (Franklin, VarSome, MobiDetails, GeneBe, InterVar, VarChat) against dual independent expert curation across diverse variant types. Discrepancies concentrate around evidence-strength weighting and near-boundary categories (LP–P, LP–VUS): concordance is highest for loss-of-function variants (mature PVS1 decision trees) but more variable for missense and splicing variants. Each tool shows its own profile (Franklin/VarSome slightly pathogenic-leaning, VarChat neutral, GeneBe higher and more variable, InterVar more conservative). The authors conclude that the tools are reliable for structured evidence but require expert adjudication for context-dependent criteria.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

A useful benchmark as laboratories prepare the transition to ACMG v4 while still operating under current recommendations. The key message — automate the structured, reserve the expert for context-dependent criteria (PM1, PP3/BP4, segregation) — is directly applicable in routine. The scope stays limited (no diagnostic-cohort metrics), but the criterion-level analysis is valuable for calibrating workflows.

Analysis by Dr Thibaut Benquey