Abstract:
|
Complex diseases such as cancer usually develop through different stages forming ordinal outcomes. Understanding the intrinsic mechanism underlying these disease stages is important for diagnosis, classification and subsequent treatment of these diseases. The etiology and development of complex diseases often involve the interactions between biomolecules, rather than individual molecules such as complicated interactions between tumor cells and immune cells in cancer immunotherapies. Predictive models are of great importance in precision or personalized medicine and other applications. In this paper, we developed a non-parametric approach to predict ordinal outcomes incorporating potentially complicated interactions between biomolecules. Simulation studies demonstrate that our approach performs well in classification, and in identification of truly informative differential pairs of predictors, when there is non-negligible interaction between predictors.
|