All-Atom GPCR-Ligand Simulation via Residual Isometric Latent Flow

TL;DR

GPCRLMD tackles the computational bottleneck of all-atom GPCR–ligand dynamics by embedding the complex into an isometric latent space via a physics-guided Harmonic-Prior VAE and then learning temporal evolution with a Residual Latent Flow. The approach decouples static topology from dynamic fluctuations, enabling parallel, long-timescale trajectory generation that preserves essential thermodynamic and kinetic properties. Evaluations against state-of-the-art baselines on GPCRMD show superior ensemble fidelity, structural validity (dihedrals and torsions), and robust ligand–receptor interactions, while dramatically improving computational efficiency (seconds per sample vs. hours). The work provides a scalable, ligand-aware framework for rapid GPCR–ligand trajectory generation with potential to accelerate structure-based drug discovery, and it lays groundwork for extending to broader protein–ligand systems.

Abstract

G-protein-coupled receptors (GPCRs), primary targets for over one-third of approved therapeutics, rely on intricate conformational transitions to transduce signals. While Molecular Dynamics (MD) is essential for elucidating this transduction process, particularly within ligand-bound complexes, conventional all-atom MD simulation is computationally prohibitive. In this paper, we introduce GPCRLMD, a deep generative framework for efficient all-atom GPCR-ligand simulation.GPCRLMD employs a Harmonic-Prior Variational Autoencoder (HP-VAE) to first map the complex into a regularized isometric latent space, preserving geometric topology via physics-informed constraints. Within this latent space, a Residual Latent Flow samples evolution trajectories, which are subsequently decoded back to atomic coordinates. By capturing temporal dynamics via relative displacements anchored to the initial structure, this residual mechanism effectively decouples static topology from dynamic fluctuations. Experimental results demonstrate that GPCRLMD achieves state-of-the-art performance in GPCR-ligand dynamics simulation, faithfully reproducing thermodynamic observables and critical ligand-receptor interactions.

Figures (19)

• Figure 1: Schematic structure of a ligand-bound GPCR. A GPCR contains seven transmembrane helices (TM1-TM7). It senses ligands and transfers this signal into the cell.
• Figure 2: Comparison of Root Mean Square Fluctuation (RMSF) between the apo (ligand-free) and complex (ligand-bound) states. The receptor exhibits significantly different fluctuation profiles when bound to the ligand (Pearson correlation $r=0.45$), demonstrating that ligand binding substantially modulates receptor dynamics. Results are shown by residue index for PDB ID $\mathrm{6LW5}$, utilizing trajectory data from aranda2025large.
• Figure 3: Pipeline of GPCRLMD. a) Framework Overview. The complete pipeline maps the GPCR-ligand trajectory $\{\mathbf{x}_i\}_{i=1}^{T-1}$ into an isometric latent space via HP-VAE, enabling the Flow Matching network to learn dynamics on a smooth, structure-preserving manifold. b) Residual Latent Flow. To effectively decouple the static equilibrium structure from stochastic dynamic fluctuations, the flow models the relative evolution $\mathbf{r}_t = \mathbf{z}_t - \mathbf{x}_0$ anchored to the initial frame $\mathbf{x}_0$. The detailed network architecture can be found in Appendix \ref{['fig:architecture_detail']}.
• Figure 4: Evaluation of ensemble fidelity and structural validity. a) Receptor and ligand RMSF profiles compared to ground-truth MD (PDBID $\mathrm{5DSG}$). b) Ligand torsional angle distributions (PDBID $\mathrm{5DSG}$). c) Coupled backbone ($\phi$-$\psi$) and side-chain ($\chi^1$-$\chi^2$) dihedral distributions (PDBID $\mathrm{5DSG}$). d) Receptor ensemble benchmarks against BioEmu and BioMD. Metrics are averaged across the test set (250 conformations per sample), evaluating only receptor coordinates to ensure fair comparison with the protein-only BioEmu baseline.
• Figure 5: Evaluation of complex dynamics and interaction stability. a) Transmembrane (TM) helix motion. Time-evolution of the critical TM2-TM6 distance (PDBID: $\mathrm{6LW5}$) alongside a structural overlay of the conformational transition. More examples in Figure \ref{['fig:tm_distance']}. b) Dynamic contact maps. Receptor-ligand and receptor-receptor interaction frequencies calculated using a dual-cutoff metric (details in Appendix \ref{['appsec:dual_contactmap_calculate']}). c) Ligand stability. Time-dependent Ligand RMSD relative to the initial frame ($t=0$). d) Free Energy Landscape (TICA). Projection of generated and reference trajectories onto the backbone energy surface, visualizing conformational space coverage. More examples and results are shown in Figure \ref{['fig:tica_bb_500ns']}, \ref{['fig:tica_sidechain_500ns']} and Table \ref{['tab:tica_js_divergence']}.