A screening model that flags high-risk variant combinations years before symptom onset.
Overview
We're building a model that screens genetic variant combinations for early-onset disease risk, aiming to flag risk years before symptoms appear.
Objectives
- Build a reproducible variant-scoring pipeline
- Validate against three independent cohorts
- Publish the scoring model as open source
Research questions
Can combinations of common variants, individually benign, predict early-onset risk when modeled jointly?
Methodology
We use gradient-boosted trees over engineered variant-interaction features, validated with nested cross-validation across cohorts.
Current progress
Model validated on two of three planned cohorts; false-positive rate within target range.
Future work
Third cohort validation, then a pre-print and open release of the scoring library.
Acknowledgements
Thanks to our volunteer cohort coordinators and the open genomics datasets that made this possible.