Laurentiu Spiridon
Affiliation
Institute of Biochemistry, Romanian Academy, Bucharest, Romania - Research
Advanced Sampling Methods for Protein–Ligand Interactions
Abstract
Introduction: Efficient and reliable sampling of protein–ligand systems remains a central challenge in molecular simulation due to the complex, high-dimensional nature of biomolecular energy landscapes and the presence of high energy barriers between metastable states. We present Generalized Coordinate Hybrid Monte Carlo (GCHMC [1]), an unbiased sampling method that combines Hamiltonian Monte Carlo dynamics with Gibbs sampling to generate statistically rigorous samples from the canonical ensemble, implemented in Robosample software package [2]. The method is formulated in internal coordinates with rigid-body representations, enabling natural treatment of constrained molecular motion and facilitating large-scale conformational changes relevant to binding processes.
Materials and Results: GCHMC alternates deterministic Hamiltonian propagation in generalized coordinates with stochastic Gibbs updates of selected degrees of freedom, ensuring detailed balance while improving exploration of configurational space. This approach allows efficient sampling of both ligand flexibility and protein–ligand relative motion without the use of biasing potentials. The internal-coordinate formulation further reduces numerical stiffness associated with bonded interactions and supports stable integration of constrained systems. The program achieves high performance through the use of efficient algorithms derived from robot mechanics.
In addition to accurate pose prediction and binding free energy estimation, the method enables computation of the full potential of mean force (PMF) along relevant binding coordinates, providing detailed characterization of the thermodynamic landscape associated with protein–ligand binding.
Conclusions: We evaluate the method on CASF diverse set of protein–ligand complexes [3] spanning different sizes, flexibilities, and interaction types. The results demonstrate robust sampling performance and consistent convergence of thermodynamic observables across systems, highlighting the potential of GCHMC as a practical and scalable framework for unbiased molecular simulation of biomolecular binding.
References
- Spiridon L, Minh DDL. “Hamiltonian Monte Carlo with Constrained Molecular Dynamics as Gibbs Sampling.” J Chem Theory Comput. 2017 Oct 10;13(10):4649-4659.
- Spiridon L, Şulea TA, Minh DDL, Petrescu AJ. “Robosample: A rigid-body molecular simulation program based on robot mechanics.” Biochim Biophys Acta Gen Subj. 2020 Aug;1864(8):129616.
- Su M, Yang Q, Du Y, Feng G, Liu Z, Li Y, Wang R. “Comparative Assessment of Scoring Functions: The CASF-2016 Update.” J Chem Inf Model. 2019 Feb 25;59(2):895-913