NERDG 2026
Poster 4 Abstract
Data Curation for Physics-Informed Cardiotoxicity Prediction via Benchmarking of hERG Pose Generation Workflows
Purvi Neema (1), Umut Ozuguzel (2), Bodhisattwa Chaudhuri (1,3,4)
(1) Department of Pharmaceutical Sciences, University of Connecticut, Storrs, CT, (2) Department of chemistry, University of Connecticut, Stamford , CT (3) Department of Chemical & Biomolecular Engineering, University of Connecticut, Storrs, CT (4) Institute of Materials Science (IMS), University of Connecticut, Storrs, CT
Presenting Author: Purvi Neema
Corresponding Author: Bodhisattwa Chaudhuri, [email protected]
Purpose
hERG blockade remains a major bottleneck in small-molecule drug development, yet the structural determinants governing ligand binding and pocket variability remain not completely understood. Cryo-EM structures (8ZYO, 8ZYN, 8ZYP, 8ZYQ) show ligand-dependent binding modes, complicating computational prediction. To address this, a computational framework was developed that integrates multiple docking techniques, advanced ligand alignment strategies, molecular dynamics (MD) simulations and MM-GBSA binding energy evaluation to systematically characterize binding determinants for a set of hERG inhibitors including astemizole, pimozide, and E4031.
Methods
Rigid and flexible workflows were performed using Schrödinger Glide and AutoDock-based methods across multiple cryo-EM receptor conformations. To understand the effect of ligand alignment, a range of ligand alignment techniques including rigid alignment like MCS-based overlays, Roshambo shape alignment, and flexible alignment technique CSAlign were applied prior to docking. Top-ranked poses underwent 10–20 ns MD simulations to evaluate stability and interaction persistence. Binding energies were assessed using Prime MM-GBSA. Pocket characteristics were quantified using SiteMap, fpocket and ligand SASA.
Results
Multiple docking techniques produced different binding orientations for each ligand, reflecting the ligand-dependent conformations seen in cryo-EM structures, with flexible docking consistently outperforming rigid protocols. Alignment methods strongly affected pose consistency across receptors: flexible and 3D shape-based overlays yielded more reproducible poses than rigid alignments. MD simulations filtered out unstable docking poses and confirmed that persistent π–π interactions with Phe656 and Tyr652 are the dominant stabilizing forces across ligands. MM-GBSA energies successfully separated high-affinity ligands from moderate ones. SiteMap and fpocket analysis showed measurable pocket expansion in certain conformations, explaining why some ligands fail to cross-dock between structures.
Conclusion
This study shows that hERG–ligand binding cannot be captured reliably using a single pose generation approach. Incorporating ligand-dependent pocket flexibility through a combination of docking, alignment, and MD refinement leads to more consistent and physically meaningful binding modes. When paired with MM-GBSA analysis, this strategy helps rationalize ligand specific interactions across different cryo-EM conformations. Overall, this workflow supports more robust data curation for physics informed prediction of hERG cardiotoxicity in early drug discovery.
Keywords
hERG channel, Cardiotoxicity prediction, Molecular docking, Ligand alignment, Molecular dynamics, MM-GBSA
Poster 4 Abstract
Data Curation for Physics-Informed Cardiotoxicity Prediction via Benchmarking of hERG Pose Generation Workflows
Purvi Neema (1), Umut Ozuguzel (2), Bodhisattwa Chaudhuri (1,3,4)
(1) Department of Pharmaceutical Sciences, University of Connecticut, Storrs, CT, (2) Department of chemistry, University of Connecticut, Stamford , CT (3) Department of Chemical & Biomolecular Engineering, University of Connecticut, Storrs, CT (4) Institute of Materials Science (IMS), University of Connecticut, Storrs, CT
Presenting Author: Purvi Neema
Corresponding Author: Bodhisattwa Chaudhuri, [email protected]
Purpose
hERG blockade remains a major bottleneck in small-molecule drug development, yet the structural determinants governing ligand binding and pocket variability remain not completely understood. Cryo-EM structures (8ZYO, 8ZYN, 8ZYP, 8ZYQ) show ligand-dependent binding modes, complicating computational prediction. To address this, a computational framework was developed that integrates multiple docking techniques, advanced ligand alignment strategies, molecular dynamics (MD) simulations and MM-GBSA binding energy evaluation to systematically characterize binding determinants for a set of hERG inhibitors including astemizole, pimozide, and E4031.
Methods
Rigid and flexible workflows were performed using Schrödinger Glide and AutoDock-based methods across multiple cryo-EM receptor conformations. To understand the effect of ligand alignment, a range of ligand alignment techniques including rigid alignment like MCS-based overlays, Roshambo shape alignment, and flexible alignment technique CSAlign were applied prior to docking. Top-ranked poses underwent 10–20 ns MD simulations to evaluate stability and interaction persistence. Binding energies were assessed using Prime MM-GBSA. Pocket characteristics were quantified using SiteMap, fpocket and ligand SASA.
Results
Multiple docking techniques produced different binding orientations for each ligand, reflecting the ligand-dependent conformations seen in cryo-EM structures, with flexible docking consistently outperforming rigid protocols. Alignment methods strongly affected pose consistency across receptors: flexible and 3D shape-based overlays yielded more reproducible poses than rigid alignments. MD simulations filtered out unstable docking poses and confirmed that persistent π–π interactions with Phe656 and Tyr652 are the dominant stabilizing forces across ligands. MM-GBSA energies successfully separated high-affinity ligands from moderate ones. SiteMap and fpocket analysis showed measurable pocket expansion in certain conformations, explaining why some ligands fail to cross-dock between structures.
Conclusion
This study shows that hERG–ligand binding cannot be captured reliably using a single pose generation approach. Incorporating ligand-dependent pocket flexibility through a combination of docking, alignment, and MD refinement leads to more consistent and physically meaningful binding modes. When paired with MM-GBSA analysis, this strategy helps rationalize ligand specific interactions across different cryo-EM conformations. Overall, this workflow supports more robust data curation for physics informed prediction of hERG cardiotoxicity in early drug discovery.
Keywords
hERG channel, Cardiotoxicity prediction, Molecular docking, Ligand alignment, Molecular dynamics, MM-GBSA
