Algorithm 2: MD-SRA

input:   Feature data X (N x P)      Target class Y (N x 1) constants: P = 11,915,233 // total number of features (SNPs)      S = 250   // reduced number of features      K = 47,660  // number of reduced models      W = 5    // number of model performance               vectors step B1: generate dense model performance matrix V for each v in W // over model performance vectors   execute step A1 of 1D-SRA algorithm   execute step A2 of 1D-SRA algorithm   transform performance matrix C by applying C[k,1:P] · C[k,P + 1]    v[1:P] = non-zero feature performance from each P of matrix C   generate matrix V by appending dense vectors v end for step B2: feature selection based on weighted elements of V define 2 clusters of matrix V based on MD-K-means clustering