Predictive Biomarker Identification for Biopharmaceutical Development

Biomarkers are the foundation of precision medicine. The identification of prognostic and predictive biomarkers is an important scientific component in advancing the drug discovery and development pipeline. Many machine learning methods have been developed to identify important prognostic biomarkers. However, most existing algorithms are not applicable for identifying predictive biomarkers because individual treatment effect is not observable. In this article, we focus on the discussion of how to modify popular ensemble learning methods and use off-the-shelf machine learning software to identify important predictive biomarker for continuous, binary, and time-to-event endpoints. We perform simulation studies to compare different methods, and we present a real example that leads to successful subgroup identification for an immunological disease treatment.