The free energy profile F(z) of a solute traversing a biological membrane can be used to estimate the rate at which the solute will translocate the membrane, which is an important property in drug development applications. We compare multiple popular methods to compute F(z) from biased molecular dynamics simulations: well-tempered metadynamics, umbrella sampling, and replica exchange umbrella sampling. We find that all three methods work well for small, uncharged solutes, but fail to converge for charged or dipolar solutes over typical simulation lengths. Such failures can be difficult to detect, as standard methods to compute uncertainties dramatically underestimate the error in systematically biased samples. We show that the committor distribution function can be used to assess whether the simulations were able to generate a valid transition state ensemble for the membrane translocation process. We also demonstrate that the pattern of accepted exchanges in replica exchange umbrella sampling calculations contains valuable information about the convergence, which is a distinct advantage of this method that is unrelated to its sampling efficiency. We use these diagnostic tools to identify the underlying physical reason for the slow convergence, which we find to be strong electrostatic interactions between charged or polar solutes and zwitterionic phospholipid head groups.

Free energy calculations of membrane translocation: Finding, overcoming, and learning from convergence bottlenecks

255th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.