About the robustness of ABF and FEB protein-ligand

From: Francesco Pietra (chiendarret_at_gmail.com)
Date: Wed Jun 28 2017 - 09:40:15 CDT

I have been dwelling for some days on applying the ABF and FEP
protein-ligand tutorial, with a protein-organic-molecule in water boxes
rather than protein-peptide. Without any restraint to any atom (except
TIP3P water), ts=1.0fs. I posted the problems in detail.

No problem was encountered with the Unbound situations, while instability
of the system raised up systematically at the first step with the Bound
situation: atoms moving too fast, usually pertaining to the ligand, less
frequently to the receptor.

I changed the setting for "r" colvar from a directly measured value on VMD
to the value resulting from "cv printframe" (I was using the last frame of
a 58.2ns unbiased MD), something that is still not wholly clear to me. All
that at no avail.

The solution was got by further decreasing the integration, to ts=0.1fs,
whereby both ABF and FEP started running (I killed the simulations as I was
using a desktop, as at the moment I have no access to a cluster, while
these simulations (or FEP alone ?) do not run on my GPU workstation.

As the Bound system had never crashed during a 58.2ns unbiased MD at
ts=1.0fs on a pure-CPU cluster, my question is: are there in general more
stringent requirements of integration than with unbiased MD when running
without any restraint on any atom?
All such problems probably arose because I am not comfortable with
restraining H-atoms because it was never proven that they are equivalent to
not restraining. Is that care irrelevant to ABF and FEP because of larger
errors elsewhere?


francesco pietra

This archive was generated by hypermail 2.1.6 : Sun Dec 31 2017 - 23:21:24 CST