Feature selection in omics prediction problems using cat scores and false nondiscovery rate control