Lessons Learned from the Calculation of One-Dimensional Potentials of Mean Force [Article v1.0]

Abstract

The origins of different computational artifacts that may occur in the calculation of one-dimensional potentials of mean force (PMF) via umbrella sampling molecular dynamics simulations and manifest as free energy offset between bulk solvent regions are investigated. By systematic studies, three distinct causes are elucidated: (i) an unfortunate choice of reference points for the umbrella distance restraint; (ii) a misfit in probability distributions between bound and unbound umbrella windows in case of multiple binding modes; (iii) artifacts introduced by the free energy estimator. Starting with a fully symmetric model system consisting of methane binding to a cylindrical host, complexity is increased through the introduction of dipolar interactions between the host and the solvent, the host and the guest molecule or between all involved species, respectively. The manifestation of artifacts is illustrated and their origin and prevention is discussed. Finally, the consequences for the calculation of standard binding free enthalpies is illustrated using the complexation of primary alcohols with α-cyclodextrin as an example.

Cell‐Penetrating Peptides: Correlation between Peptide‐Lipid Interaction and Penetration Efficiency

“…One possible reason for the slow PMF convergence in our work could be due to only distance restraint between the centre of mass (COM) of the peptide and the COM of the lipid bilayer (and not the orientation restraint of the peptide) was used during the umbrella sampling simulations. This agrees with a recent study by Markthaler et al [16] . who found that the standard umbrella sampling simulation of 1‐dodecanol binding to α‐cyclodextrin with orientation restraint on 1‐dodecanol only requires 20 ns simulations per umbrella window to achieve convergence, while the same simulation with only distance restraint between the COMs requires 140 ns per umbrella window for convergence.…”

“…One possible reason for the slow PMF convergence in our work could be due to only distance restraint between the centre of mass (COM) of the peptide and the COM of the lipid bilayer (and not the orientation restraint of the peptide) was used during the umbrella sampling simulations. This agrees with a recent study by Markthaler et al [16] who found that the standard umbrella sampling simulation of 1-dodecanol binding to α-cyclodextrin with orientation restraint on 1-dodecanol only requires 20 ns simulations per umbrella window to achieve convergence, while the same simulation with only distance restraint between the COMs requires 140 ns per umbrella window for convergence. Although it is tempting to include orientation restraint on our peptides in the umbrella sampling simulations, it is, however, difficult to determine which initial orientation of the peptide relative to the lipid bilayer to be used, as different initial orientation may lead to different interaction energy with the lipid bilayer and hence its PMF value.…”

“…PMF convergence issue can be due to various reasons. This include (1) insufficient overlap between the neighbouring umbrella windows, (2) insufficient sampling time, or (3) computational artefacts from the improper choice of reference points for the umbrella distance restraint, a misfit in probability distributions between the bound and unbound umbrella windows, or the offsets introduced by the UI estimator due to non‐Gaussian shaped biased distribution functions [16] . As for our work, the PMF convergence issue is more likely to be due to insufficient sampling time rather than the other two aforementioned reasons.…”

On the Effect of the Various Assumptions and Approximations used in Molecular Simulations on the Properties of Bio‐Molecular Systems: Overview and Perspective on Issues

“…The application to different non‐bonded force distance ranges is rather stable and relatively insensitive to the time between updates of the long‐ranged forces [78,79] . Of course, how well this approximation functions also depends on the system simulated and the property (thermodynamic, structural, dynamic, dielectric) of interest [57,58,61,62,78–80] . For an overview of the types of MTS algorithms we refer to [79] …”

Binding free energies for the SAMPL8 CB8 “Drugs of Abuse” challenge from umbrella sampling combined with Hamiltonian replica exchange

“…Among the many possible methodological variants to compute binding free energies for host-guest systems within a rigorous thermodynamic framework [1], umbrella sampling (US) [2] has emerged as an easy-to-use approach that is supported by many publicly available pre-and post-analysis tools [3]. If certain artefacts related to insufficient sampling are accounted for [4], the method yields robust estimates. Based on our previous experience with this approach in the context of cyclodextrin host-guest systems [4], we aimed for the evaluation of the simulation protocol in the SAMPL8 challenge, featuring more complex guest molecules than considered before.…”