Robust Automated\nTruncation Point Selection for Molecular\nSimulations

Quantities calculated from molecular simulations are\noften subject\nto an initial bias due to unrepresentative starting configurations.\nInitial data are usually discarded to reduce bias. Chodera’s\nmethod for automated truncation point selection [J. Chem. Theory Comput.\n2016, 12, 4, 1799–1805] is popular but has not been\nthoroughly assessed. We reformulate White’s marginal standard\nerror rule to provide a spectrum of truncation point selection heuristics\nthat differ in their treatment of autocorrelation. These include a\nmethod effectively equivalent to Chodera’s. We test these methods\non ensembles of synthetic time series modeled on free energy change\nestimates from long absolute binding free energy calculations. Methods\nthat more thoroughly account for autocorrelation often show late and\nvariable truncation times, while methods that less thoroughly account\nfor autocorrelation often show early truncation, relative to the optimal\ntruncation point. This increases variance and bias, respectively.\nWe recommend a method that achieves robust performance across our\ntest sets by balancing these two extremes. None of the methods reliably\ndetected insufficient sampling. All heuristics tested are implemented\nin the open-source Python package RED (github.com/fjclark/red).