What Limits a Reference Rate: The State Definition, Not the Estimator
Year: 2026
Type: article
Abstract: Enhanced-sampling methods are validated against reference rates from long unbiased simulations, and the reference is treated as the answer. We show that the reference carries an accuracy floor set by an analysis choice that is almost never reported. Using 39.2 μs of alanine dipeptide in explicit water, in 16 replicas totaling 196 million analyzed frames, we apply two accepted rate estimators to two conventional definitions of the same conformational transition in a full two-by-two design. Given matched states, the Hill relation and core-set milestoning agree to between 0.08 and 0.56%. Given matched estimators, two ordinary definitions of the same transition, angular cores and a distance-difference threshold, disagree by 6.4 to 7.0%. The offset does not diminish as sampling increases from 15.0 to 39.2 μs. It is negligible on a transition 721 to 769 times slower depending on which definition sets the comparison, where the two definitions select event sets differing by a single count, and a Markov state model cannot arbitrate because no lag is simultaneously long enough for Markovianity and short enough to leave a 143 ps mean first passage time undistorted. Reference rates should be published with their state definitions.
Institution: University of Wyoming
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7203724799
