Docking‑based virtual screening of BRD4 (BD1) inhibitors: assessment of docking methods, scoring functions and in silico molecular properties

Abstract

To enhance the accuracy of virtual screening for bromodomain-containing protein 4 (BRD4) inhibitors, two docking protocols and seven scoring functions were compared. A total of 73 crystal structures of BRD4 (BD1) complexes were selected for analysis. Firstly, docking was carried out using both the LibDock and CDOCKER methods. The CDOCKER protocol was shown to be more effective based on the root mean square deviation (RMSD) values (in Å) between the docking positions and the co-crystal structures, achieving a docking accuracy rate of 86.3%. Then, among the various scoring functions (LigScore1, LigScore2, PLP1, PLP2, PMF, PMF04 and Ludi3), PMF showed the highest correlation with inhibition constants (r² = 0.614), while Ludi3 scored lowest (r² = 0.266). Finally, using ligand descriptors from PubChem, a strong correlation (r² > 0.5) with inhibition constants for heavy atom count was found. Based on these comprehensive evaluations, the PMF scoring function emerged as the best tool for docking-based virtual screening of potential BRD4 (BD1) inhibitors. And the correlation between molecular properties and BRD4 (BD1) ligands also provided information for future design strategies.