|
Abstract:
|
For personalized medicine, goals are choose important biomarkers to accurately predict treatment outcomes and not to choose unimportant biomarkers. For variable selection, the lasso and the elastic net have yielded promising results. However, selecting the right amount of penalization is critical to achieving the two goals. For this, cross-validation (CV) is commonly used. It tends to provide high prediction accuracy and a high true positive rate, at the cost of too many false positives. Stability selection (SS) controls the number of false positives, but at the cost of selecting too few true positives. We propose prediction-oriented marker selection (PROMISE), which combines SS with CV to include the advantages of both methods. Our application of PROMISE with the lasso and elastic net in data analyses show that PROMISE produces a more sparse solution than CV, reducing the false positives compared to CV, while giving similar prediction accuracy and true positives. The performance of SS varies according to the data. PROMISE can be applied in many fields to select regularization parameters when the goals are to minimize false positives and maximize prediction accuracy.
|
