Title: Making X-ray and Cryo-EM Heterogeneity Actionable: Physics-Based, Density-Driven Ensemble Refinement for Modern Therapeutic Modalities and AI/ML-Ready Datasets

Conference Registration: https://www.industrialbiostructuresamerica.org/

Abstract:
Structure-guided drug discovery increasingly relies on X-ray crystallography and cryogenic electron microscopy (Cryo-EM) structures to interpret binding hypotheses, guide lead optimization, and prioritize compounds. Yet many pharmaceutically relevant systems exhibit intrinsic disorder, partial occupancy, and heterogeneous ligand binding that are often compressed into a single-conformer model. This simplification can obscure key decisions—such as which ligand pose is supported by density, which protonation or tautomeric state drives the interaction network, and whether apparent ligand strain reflects true bound-state physics or modeling artifacts. These challenges are especially acute for modern therapeutic modalities, including macrocycles, cyclic peptides, proteolysis-targeting chimeras (PROTACs), covalent inhibitors, and metal- containing ligands, where multiple coexisting states and correlated protein–ligand rearrangements frequently determine mechanism.

We present an application-focused workflow for disorder-aware ensemble refinement that is physics-based and density-driven, producing compact ensembles that retain experimentally supported heterogeneity rather than forcing a single averaged structure. The approach explicitly represents alternative conformational states, presence/absence behavior, and heterogeneous binding scenarios (including fragment mixtures and cases where different ligands occupy the same site). By combining real-space density targets with chemically rigorous refinement, the method supports direct determination of discrete chemical and stereochemical hypotheses—including alternative protomer and tautomer states, ring/flip states, and chiral states—within the same refinement framework used to evaluate competing binding modes and conformers.

Across case studies spanning challenging chemotypes and modalities, disorder-aware ensemble refinement improves practical model quality relative to deposited structures, including reduced steric clashes, improved stereochemical quality, improved density agreement, and reduced artifactual ligand strain. The resulting ensembles also provide more reliable starting points for downstream sampling and clustering workflows, supporting molecular dynamics simulations and enhanced sampling methods such as replica exchange and metadynamics. Finally, by preserving disordered loops, alternate binding modes, and chemically meaningful state assignments that are often underrepresented in curated structural corpora, these ensembles produce higher-quality structural datasets for artificial intelligence and machine learning (AI/ML) model development—improving both training coverage and evaluation realism for models that depend on accurate representations of conformational diversity and binding heterogeneity.

This presentation will emphasize practical impact in drug discovery, highlighting how explicitly modeling heterogeneity with a physics-based, density-driven, chemically rigorous refinement workflow improves decision-making for modern modalities and provides AI/ML-ready structural inputs that better reflect the true diversity of experimentally observed states.