a) Background: Current challenges in prognosis of prostate cancer using gene expression signatures. b) Materials and Methods: TCGA training set and six independent test sets; BLUP-HAT method is applied to transcriptomic data and multi-omics data for the selection of prognostic markers. c) Results: Comparison of prediction performances between various prognostic models derived using BLUP-HAT method as well as the models including the genes in three commercial prognostic panels. d) Conclusion and Discussion: Clinical significance and application of the new method and the current limitations.