Gravar-mail: A dropout-regularized classifier development approach optimized for precision medicine test discovery from omics data