An integrated machine learning and computational framework with experimental validation for the identification of novel CXCR4 inhibitors.

Chemokine receptor 4 (CXCR4) is a clinically significant G protein-coupled receptor implicated in HIV-1 entry, cancer progression, immune regulation, and metastatic dissemination, making it an attractive therapeutic target. This study employed an integrated computational and experimental framework to identify novel small-molecule CXCR4 inhibitors. A curated dataset of 608 compounds from peer-reviewed literature and patents was used to train machine-learning classification models. Decision Tree, Logistic Regression, and AdaBoost models showed balanced performance across key metrics, and external validation on 2146 in-house compounds identified 44 consensus CXCR4 inhibitors. Molecular docking analyses suggested favorable binding modes and key interactions comparable to those predicted for the reference inhibitor IT1t. One hundred-nanosecond molecular dynamics simulations indicated stable CXCR4-ligand complexes, with equilibration occurring within approximately 20 ns and backbone RMSD values maintained between 4 and 8 Å. MM/GBSA free-energy calculations demonstrated favorable energetics, with IS00622 exhibiting the strongest affinity (-70 kcal/mol), followed by IT1t, IS00998, and IS00179. In vitro assays identified IS00127 as a promising lead, showing strong antiproliferative activity against MDA-MB-231 cells and minimal toxicity toward HEK293 cells. ELISA assays confirmed dose-dependent CXCR4 downregulation with negligible effects on CXCR7, indicating high functional selectivity. Overall, this integrative strategy accelerates the discovery of potent, selective CXCR4 inhibitors for translational research.