Development of random subsample classifiers identifying mechanism-specific subgroups of type 2 diabetes

Abstract

Recently, precision medicine research on type 2 diabetes has been actively conducted, but currently, studies on clustering of patient groups using clinical variables are dominant, and stratification of type 2 diabetes patient groups based on disease mechanisms is still difficult. In this study, to solve this problem, the mechanisms of type 2 diabetes were first defined and hierarchically structured to enable bioinformatics analysis. In addition, I developed the S-cube, a method for discovering the patient groups and their corresponding disease mechanisms from a limited number of transcriptome data. S-cube is a method of discovering each patient group and mechanism by creating combinations of various patients and discriminating data through expert-based prediction principles. As a result of comparing the classification performance with existing methods such as SVM and Pasting, it was confirmed that the S-cube showed higher classification performance. In addition, 21 patient groups were successfully discovered by applying the S-cube method to the type 2 diabetes pancreatic islet transcriptome data. Finally, a surrogate experimental model was constructed to predict drug response in the discovered patient group. The series of procedures including S-cube proposed in this study can be applied not only to type 2 diabetes but also to other diseases. Therefore, it is expected that it will be able to contribute to the application of disease mechanism-based precision medicine of other diseases like rare diseases.