Enhanced Sampling for Pharmaceutical Molecular Dynamics: A Systematic Review of Accuracy, Efficiency, Reproducibility, and Mechanistic Validity

Molecular dynamics can characterize conformational ensembles, energy landscapes, ligand-recognition pathways, allosteric transitions, and molecular mechanisms relevant to pharmaceutical research, but conventional trajectories frequently undersample rare events and slow collective motions. Enhanced-sampling methods address this limitation through biased potentials, generalized ensembles, replica exchange, pathway sampling, adaptive trajectory generation, or learned collective variables. This systematic methodological review evaluates how these approaches support four distinct claims: accurate recovery of equilibrium or kinetic properties, computational efficiency, reproducibility across implementations, and mechanistic validity. A protocol-driven search and selection process was organized around explicit review questions, reproducible database syntax, prespecified eligibility criteria, duplicate handling, structured extraction, and qualitative appraisal. The evidence was synthesized by method family, target observable, estimator requirements, implementation context, convergence assessment, and validation stage. The reviewed literature indicates that acceleration is not equivalent to correctness: apparent exploration, convergence, or computational speed may coexist with collective-variable omission, force-field error, reweighting instability, implementation sensitivity, or distorted kinetics. Collective-variable and machine-learning approaches can improve access to slow modes, whereas replica, tempering, pathway, and adaptive methods offer complementary strategies for overcoming barriers or constructing transition ensembles. However, comparative performance remains conditional on the molecular system, observable, initial information, algorithmic settings, and validation reference. Reproducibility further depends on transparent software versions, parameters, randomization, input files, analysis definitions, and uncertainty assessment. Mechanistic interpretation is strongest when sampled ensembles and pathways are consistent with independent experimental observables and when alternative explanations are examined. Enhanced sampling may therefore support pharmaceutical molecular dynamics when method choice, estimand, convergence, uncertainty, and validation are aligned, but no method family can be considered universally accurate, efficient, reproducible, or mechanistically valid.