Schematic illustration of the analysis strategy. The initial dataset was randomly divided into training and test sets. In a training set, genes below the threshold p-value in a pathway were subjected to PCA. Two models were constructed, specifically, one principal component model using the most significantly associated principal component and weighted model using multiple principal components, and evaluated in the test set. This procedure was repeated 1,000 times with random training and test sets. Finally, median values of statistics from test sets were measured to select significant pathways.