Title: Accurate Binding Free Energy Calculations with MovableType and Density-Driven Structural Preparation

Conference Information: https://www.feworkshop.org/

Abstract:

Accurate absolute binding free energy calculations require not only efficient sampling strategies but also complete and chemically accurate input structures. We describe how the MovableType (MT) framework, in combination with advanced density-driven structure preparation, improves prediction accuracy, model robustness, and computational efficiency in free energy workflows.

MovableType divides conformational sampling into local and global regimes. Local sampling (“MT proper”) decomposes inter- and intramolecular energies into pairwise potentials, enabling efficient and statistically meaningful sampling of local conformational space. Global sampling incorporates docking, rotamer enumeration, enhanced MD, and full MD/MC simulations to capture large-scale flexibility. With the integration of OpenMM as an optional MD engine in the DivCon Discovery Suite, enhanced MD will soon be used directly within MT workflows for added flexibility and precision.

A key component of accurate free energy modeling is high-quality input structures. We introduce an automated, density-guided structure preparation protocol for rebuilding disordered regions—flexible loops, mobile rotamers, and protonation/tautomeric states—using experimental X-ray or Cryo-EM maps. Implemented in DivCon, this method performs φ/ψ-guided loop generation, side-chain rotamer optimization, and protonation state sampling, all refined in real space using gradients derived from QM, MM, or hybrid QM/MM Hamiltonians. The Z-score of the difference density (ZDD) is applied at each stage—backbone, side-chain, and protonation selection—to ensure agreement with experimental data.

These chemically complete models improve downstream MD stability, preserve critical loop–ligand interactions, and enhance binding free energy predictions using MovableType. Case studies demonstrate improved accuracy and reduced variability, supporting the broader goal of reliable and scalable free energy modeling. This work also enhances structural fidelity in AI/ML workflows, where structural quality directly impacts predictive performance.