Leveraging a Deep Learning Model and Computational Screening for the Discovery of ABCG2 Inhibitors against Drug-Resistant Breast Cancer.
TL;DR
This research identified a small-molecule compound, F0161-0219, as a promising ABCG2 inhibitor, which led to a 1.94-fold sensitization enhancement in ADR-resistant cells, and underscores the potential of leveraging deep learning models and computational screening for the discovery of ABCG2 inhibitors.
Abstract
Multidrug resistance (MDR) mediated by ATP-binding cassette transporter G2 (ABCG2) remains a major clinical challenge in breast cancer chemotherapy, primarily due to its role in reducing intracellular adriamycin (ADR) concentrations. To address this challenge, we developed an integrated computational strategy for the rapid identification of potent ABCG2 inhibitors capable of reversing ADR resistance. We first established the Prediction of Protein-Ligand's pKd (PPLK), a specialized deep learning model that employs multimodal sequence-ligand representations and feedforward neural networks to accelerate the discovery of ABCG2 inhibitors. Next, computational virtual screening of a compound library (∼1.58 million) was conducted, coupled with molecular dynamics simulations, focusing on the key binding residues THR542, PHE439, ASN436, and SER440. Among the top-ranked candidates, reserpine was validated as a potent ABCG2 inhibitor, achieving 1.90-fold sensitization in ADR-resistant cells, thereby confirming the predictive accuracy of the PPLK model. Notably, we identified a small-molecule compound, F0161-0219, as a promising ABCG2 inhibitor, which led to a 1.94-fold sensitization enhancement in ADR-resistant cells. Direct target engagement of F0161-0219 with ABCG2 was confirmed by biolayer interferometry, DARTS, and CETSA assays. Collectively, our research underscores the potential of leveraging deep learning models and computational screening for the discovery of ABCG2 inhibitors.