Overview
- Analyzes GRAfT clinical trial data containing microRNA panels and heart transplant rejection outcomes.
- Builds repeated sample-splitting and penalized logistic regression pipelines to identify stable biomarkers.
- Develops Neyman-Pearson classifiers for ACR vs. non-ACR and AMR vs. non-AMR with controlled false positive rates.
Methods and contribution
The project focuses on high-stakes clinical classification where false positives must be
tightly controlled. Instead of treating feature selection as fixed, it explicitly studies how
repeated splitting and model instability affect downstream classification reliability.
The workflow combines biomarker discovery, uncertainty assessment, and Neyman-Pearson style
classification to support interpretable and robust clinical decision tools.
Materials
Paper
Slides
Clinical data restricted