PIM-1 kinase is a serine-threonine phosphorylating enzyme with implications in multiple types of malignancies, including prostate, breast, and blood cancers. Developing better search methodologies for PIM-1 kinase inhibitors may be a good strategy to speed up the discovery of an oncological drug approved for targeting this specific kinase. Computer-aided screening methods are promising approaches for the discovery of novel therapeutics, although certain limitations should be addressed. A frequent omission that is encountered in molecular docking is the lack of proper implementation of scoring functions and algorithms on the post-docking results, which usually alters the outcome of the virtual screening. The current study suggests a method for post-processing docking results, expressed either as binding affinity or score, that considers different binding modes of known inhibitors to the studied targets while making use of in vitro data, where available. The docking protocol successfully discriminated between known PIM-1 kinase inhibitors and decoy molecules, although binding energies alone were not sufficient to ensure a successful prediction. Logistic regression models were trained to predict the probability of PIM-1 kinase inhibitory activity based on binding energies and the presence of interactions with identified key amino acid residues. The selected model showed 80.9% true positive and 81.4% true negative rates. The discussed approach can be further applied in large-scale molecular docking campaigns to increase hit discovery success rates.
Keywords: PIM-1 inhibitors; PIM-1 kinase; amino acid interactions; data clustering; logistic regression; predictive score; protein kinase inhibitors; virtual screening.