Conformational Entropies and Order Parameters: Convergence, Reproducibility, and Transferability

Conformational entropy provides major contributions to protein folding and functions, such as ligand binding, making it a potentially important driver of biologically relevant processes. NMR spectroscopy is a unique technique to estimate conformational entropy changes at atomic resolution, an approach that can be favorably augmented by comparisons with results from molecular dynamics (MD) simulations, for example, by generating an order-parameter-to-entropy dictionary. Here, we address critical issues pertaining to such an approach, including reproducibility, convergence, and transferability by analyzing long (380 ns -1 ms) MD trajectories obtained for five different proteins. We observe that order parameters and conformational entropies calculated over 10-100 ns windows are typically well converged among individual MD trajectories and reproducible between pairs of independent trajectories, when calculated on a per-residue level. However, significant discrepancies sometimes arise for the total conformational entropy evaluated as the sum of the residue-specific entropies, especially in cases that involve rare transitions to alternative conformational states. Furthermore, we find that the order-parameter-to-entropy dictionary depends strongly on the protein and the sampling frequency, but much less so on the molecular dynamics force field. Thus, the transferability of the dictionary is poor between proteins but relatively good between different states (e.g., different ligand-bound complexes) of the same protein, provided that a protein-specific dictionary has been derived