The calculation of the free Gibbs energy of transferring a molecule from the aqueous phase to octanol phase (or vice versa) by using umbrella sampling in order to estimate the oil-water partition coefficient (logP) seems to be a process that strongly depends on the chosen umbrella windows. Thus, if you repeat the molecular dynamics to obtain new umbrella windows and choose different configurantions, you obtain different potential of mean forces, resulting in different free energies and different logP. Has anyone investigated this methodology in terms of reproducibility? Please, let me know about it.