One-sided Kernel Principal Component Analysis

Abstract

Principal component analysis (PCA) is widely used in omics research, such as metabolomics. Kernel principal component analysis (KPCA) is also widely used in machine learning because it can compute various nonlinear PCA depending on the flexible setting of the kernel function, but it is rarely used in omics research. One of the reasons why KPCA has not been used for omics data analysis is that it cannot directly calculate principal component loadings to select important variables such as metabolites. In this study, we propose one-sided KPCA that can directly compute and use principal component loadings to select important variables.