Pathway and system response elucidation in non-equilibrium molecular dynamics simulations

Protein-ligand binding and unbinding are fundamental to biological function and drug action. Molecular dynamics (MD) simulations, particularly dissipation-corrected targeted molecular dynamics (dcTMD), offer a route to reconstruct free-energy and friction profiles from non-equilibrium pulling trajectories, enabling the prediction of binding and unbinding rates when combined with temperature-boosted Langevin simulations. However, dcTMD rests on key theoretical assumptions and is sensitive to practical simulation choices. This thesis investigates the accuracy and limitations of dcTMD from complementary angles. First, we examine its theoretical foundation using an exact generalized Langevin equation system under external driving, identifying a spatially structured non-equilibrium response force whose behaviour depends on the pulling regime. Second, we assess the sensitivity of dcTMD to the simulation setup in the trypsin-benzamidine complex, showing that the hydration state -- particularly buried water molecules in the binding pocket -- exceeds the effects of force field choice on the free-energy estimate. Third, we address the breakdown of the cumulant approximation of Jarzynski's equation underlying dcTMD, when multiple unbinding pathways coexist by introducing a pathway-separation workflow based on time-series clustering. This enables path-wise free-energy and non-equilibrium friction estimates and yields kinetic rate predictions in closer agreement with experiment. Finally, we integrate simulations with stereographic atomic-force microscopy to show that pathway-dependent friction is a physically measurable property. Together, these results define the conditions under which dcTMD provides reliable predictions: slow pulling aligned with the intrinsic unbinding timescale, careful control of hydration, and separation of distinct unbinding pathways. The developed pathway-separation and analysis pipeline, alongside recommendations for system preparation and benchmarking, advances dcTMD-based studies of protein-ligand systems.