1. Introduction
We have covered the history, the rules, the data, and the pitfalls. But the field of variant classification is not static. We are on the cusp of a revolution driven by two forces: high-throughput functional genomics and artificial intelligence.
2. MAVE: Multiplexed Assays of Variant Effect
Historically, functional assays were done one variant at a time. MAVE (or "saturation mutagenesis") changes this by testing every possible amino acid substitution in a gene simultaneously in a single experiment.
Projects like the Atlas of Variant Effects Alliance aim to map the functional impact of all human variants.
Impact: Instead of waiting for a patient to present with a VUS, we pre-calculate the functional score for every possible VUS. When the patient arrives, the data is already there (PS3).
3. The AI Revolution
Deep learning is transforming in-silico prediction from a "supporting" line of evidence to a potential "strong" one.
AlphaMissense
Developed by Google DeepMind. It adapts the AlphaFold protein structure model to predict pathogenicity. It achieves state-of-the-art performance by understanding the 3D structural context of the variant.
PrimateAI-3D
Developed by Illumina. It uses natural selection in non-human primates as a training set. Since primates are our closest relatives, variants tolerated in them are likely benign in us.
4. ACMG v4.0 Preview
The next iteration of the guidelines is currently in development. While not yet released, discussions in the community point to major shifts:
- Fully Bayesian: Moving away from the point system approximation to a true Bayesian calculation.
- Continuous Data: Allowing continuous scores from AI tools rather than binning them into "Supporting" or "Strong".
- Simplification: Reducing the number of criteria by collapsing redundant ones.
5. Conclusion: The Human in the Loop
Despite these advances, the role of the human curator remains vital. AI can predict, but it cannot diagnose. It cannot talk to a patient, understand a complex family history, or make the ethical decision to report a finding that will change a life.
The future of variant classification is AI-augmented curation, not AI-replaced curation.