Prediction-Oriented Marker Selection (PROMISE): With Application to High-Dimensional Regression.

Area of Science:

• Biostatistics
• Genomics
• Personalized Medicine

Background:

• Biomarker selection is crucial for personalized medicine to predict treatment success.
• High-dimensional genomic data presents challenges in identifying truly predictive biomarkers.
• Penalized regression methods like lasso and elastic net are promising but require careful parameter tuning.

Purpose of the Study:

• To develop a novel variable selection method, PROMISE, that balances prediction accuracy with the control of false positives.
• To address the limitations of standard cross-validation (CV) and stability selection (SS) in biomarker discovery.
• To improve the efficiency and reliability of biomarker selection in high-dimensional genomic data.

Main Methods:

• PROMISE integrates stability selection (SS) with cross-validation (CV) to optimize regularization parameter selection.
• The method was applied using lasso and elastic net penalized regression techniques.
• Performance was evaluated against standard CV and SS approaches.

Main Results:

• PROMISE yields sparser solutions with fewer false positives compared to CV.
• PROMISE maintains good prediction accuracy while reducing type I and type II errors.
• PROMISE demonstrates superior prediction accuracy and true positive rates compared to SS.

Conclusions:

• PROMISE offers a robust approach for biomarker selection in personalized medicine.
• The method effectively minimizes false positives and maximizes prediction accuracy.
• PROMISE is applicable across various fields requiring regularization parameter selection.