Identification of BRD4 (BD1) Selective Inhibitors Based on Machine Learning, Molecular Docking and Molecular Dynamics Simulation

Abstract Bromodomain‐containing protein 4 (BRD4) is regarded as a promising therapeutic target for cancer and inflammation due to its crucial role in regulating transcription elongation and gene expression. However, current pan‐BRD4 inhibitors have shown resistance and dose‐limiting toxicity in clinical trials, highlighting the need for domain selective BRD4 inhibitors. To identify domain selective BRD4 inhibitors, we constructed the machine learning models with good predictive performance based on reported inhibitors of BRD4 first bromodomain (BD1) and second bromodomain (BD2). Then, the optimal models and molecular docking were then employed to screen for compounds with potential BRD4 (BD1) inhibitory activity from the ZINC database. Afterward, three hit compounds were selected for further evaluation through molecular dynamics simulations. The results indicated that the binding of the three hit compounds with BRD4 (BD1) stabilized the protein structure during the 250 ns of molecular dynamics simulations. In comparison, the binding of these compounds with BRD4 (BD2) did not exhibit the same stabilizing effect as BRD4 (BD1), indicating their potential selective inhibition of BRD4 (BD1) over BRD4 (BD2). This study demonstrates the effectiveness of machine learning, molecular docking, and molecular dynamics simulation for the discovery of selective inhibitors, and offers novel insights into the research of BRD4 inhibitors.