Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GeoDirDock: Guiding Docking Along Geodesic Paths

Raúl Miñán, Javier Gallardo, Álvaro Ciudad, Alexis Molina

TL;DR

GeoDirDock introduces geodesic-guided diffusion for molecular docking, directing the denoising process along shortest-path trajectories in the translation-rotation-torsion space $P=\mathbb{T}^3\times SO(3)\times SO(2)^m$ to target binding regions. By incorporating expert guidance through a guiding vector $V_{guide}$ and a gamma-controlled update, the method achieves higher RMSD accuracy and more physically plausible poses than blind docking, while enabling angle-transfer during maximal common substructure docking for lead optimization. The approach demonstrates robust improvements across Apo/Holo docking and shows promising generalization in MCS benchmarks (e.g., BACE/D3R4). Future work will aim at broader chemical generalization, protein flexibility, and improved priors to further reduce steric clashes and enhance prospective docking performance.

Abstract

This work introduces GeoDirDock (GDD), a novel approach to molecular docking that enhances the accuracy and physical plausibility of ligand docking predictions. GDD guides the denoising process of a diffusion model along geodesic paths within multiple spaces representing translational, rotational, and torsional degrees of freedom. Our method leverages expert knowledge to direct the generative modeling process, specifically targeting desired protein-ligand interaction regions. We demonstrate that GDD significantly outperforms existing blind docking methods in terms of RMSD accuracy and physicochemical pose realism. Our results indicate that incorporating domain expertise into the diffusion process leads to more biologically relevant docking predictions. Additionally, we explore the potential of GDD for lead optimization in drug discovery through angle transfer in maximal common substructure (MCS) docking, showcasing its capability to predict ligand orientations for chemically similar compounds accurately.

GeoDirDock: Guiding Docking Along Geodesic Paths

TL;DR

GeoDirDock introduces geodesic-guided diffusion for molecular docking, directing the denoising process along shortest-path trajectories in the translation-rotation-torsion space to target binding regions. By incorporating expert guidance through a guiding vector and a gamma-controlled update, the method achieves higher RMSD accuracy and more physically plausible poses than blind docking, while enabling angle-transfer during maximal common substructure docking for lead optimization. The approach demonstrates robust improvements across Apo/Holo docking and shows promising generalization in MCS benchmarks (e.g., BACE/D3R4). Future work will aim at broader chemical generalization, protein flexibility, and improved priors to further reduce steric clashes and enhance prospective docking performance.

Abstract

This work introduces GeoDirDock (GDD), a novel approach to molecular docking that enhances the accuracy and physical plausibility of ligand docking predictions. GDD guides the denoising process of a diffusion model along geodesic paths within multiple spaces representing translational, rotational, and torsional degrees of freedom. Our method leverages expert knowledge to direct the generative modeling process, specifically targeting desired protein-ligand interaction regions. We demonstrate that GDD significantly outperforms existing blind docking methods in terms of RMSD accuracy and physicochemical pose realism. Our results indicate that incorporating domain expertise into the diffusion process leads to more biologically relevant docking predictions. Additionally, we explore the potential of GDD for lead optimization in drug discovery through angle transfer in maximal common substructure (MCS) docking, showcasing its capability to predict ligand orientations for chemically similar compounds accurately.
Paper Structure (25 sections, 20 equations, 10 figures, 5 tables, 3 algorithms)

This paper contains 25 sections, 20 equations, 10 figures, 5 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related work
  3. Methods
  4. Boundaries and Geodesics
  5. Soft Constraints
  6. Evaluation
  7. Experiments
  8. Evaluation of docking poses
  9. Physical plausibility of informed diffusion poses
  10. Angle transfer as maximal common substructure docking
  11. Conclusions and Future Work
  12. Algorithms and computations
  13. Geodesics in $\mathbb{T}^3$
  14. Geodesics in $SO(3)$
  15. Geodesics in $SO(2)^m$
  16. ...and 10 more sections

Figures (10)

  • Figure 1: Guidance in $\mathbb{T}^3$ for a sample sphere in 1a0q.
  • Figure 2: Rotation state vectors in $S^2$ for a reflected molecule versus its ground truth.
  • Figure 3: $SO(3)$ guidance in $S^2$ for a sample rotation state.
  • Figure 4: Guidance in $SO(2)^m$ for a 2 torsion angle compound.
  • Figure 5: Mean and median distance to center of sphere versus denoising step.
  • ...and 5 more figures