ACMG classification shouldn’t be a black box

Automated ACMG classification is now standard. But most engines output a single word — Pathogenic, VUS, Benign — and hide the reasoning that produced it. For a clinical geneticist who has to sign, that is a problem: you cannot defend a classification you cannot see.

From rules to points

The 2015 ACMG/AMP framework combined criteria through a set of qualitative rules. The ClinGen Sequence Variant Interpretation (SVI) working group later reframed it as a Bayesian point system: each criterion contributes evidence weight, and the sum maps to a classification with a defined probability. It is more transparent and more consistent.

Recording what wasn’t applied

Innovare’s engine is aligned with those recommendations — and adds something most do not: it logs an auditable ledger of every criterion it evaluated, including the ones it deliberately did not apply. A criterion that was considered and rejected is as informative as one that fired.

Review the reasoning, then sign

The output is not a verdict to accept on faith. It is a structured argument: which criteria fired, at what strength, which were excluded and why, and how they summed to a classification. The geneticist reviews the how, adjusts if their clinical judgment differs, and signs. The machine proposes; the expert decides.