PRK1 has been identified as potential drug target for prostate cancer therapy. Due to the absence of crystal structure, multiple PRK1 homology models were generated. An in-house library of compounds tested on PRK1 was docked into generated homology models. In most cases the correct pose of the inhibitor could be identified by ensemble docking, while there was still a challenge of finding a reasonable scoring function able to rank compounds according to their biological activity. Thus, we estimated the binding free energy using the MMPB(GB)SA and QM/MM-GBSA methods after energy minimization in implicit solvent. The QSAR model was designed on the basis of binding free energy scores in order to predict the pIC50 values of compounds, showing a good correlation with experimental data. The prospective validation of this approach on two external datasets identified a number of highly potent PRK1 inhibitors, proving the outstanding performance of the method. Finally, recently released crystal structures of PRK1 were compared to homology models, showing a high similarity of their structures, especially at the inhibitor binding site.