[AMBER] Absolute Binding Free Energy of Protein+Ligand Complex in AMBER20 - Boresch restraints, Smoothstep

From: Josiah Bones <josiah.bones.anu.edu.au>
Date: Fri, 18 Jun 2021 06:29:58 +0000

Hi there,

I asked a question about some parts of this elsewhere, but I realised I needed to simplify my previous question into something a bit more basic and address some issues in a single location that I can't find mentioned in any tutorials.

I am wondering how to successfully complete what I believe is a fairly standard ABFE simulation in the context of AMBER20 on GPU. The goal is to capture the absolute binding of a ligand to a protein from a high-quality crystal structure.

I believe I need 3x simulation phases +1 post simulation calculation to correctly capture the dG.

  1. Complex+Ligand: Simulate turning on Boresch restraints.
     * I have defined these restraints in a DISANG file based on a high quality crystal structure using general recommendations mentioned elsewhere, but can't find any clue as to the correct way to simulate this leg anywhere, and how the .mdin file/s should look.
     * Is this done without softcore potentials?
     * What do the lambda windows look like, and how should this change with reference to the restraint file?
     * Does the restraint file change between windows?
     * What is V0 and V1 in this context?
     * What other key .mdin toggles need attention beyond the defaults? Is dvdl_norest significant here, despite being deprecated?
     * I would be absolutely thrilled to get commentary in particularly for this step, as it is completely elusive to me, and I have no idea how to solve it. Comments in a recent 2020 paper of TI changes in Amber made me assume that this was now essentially automated in Step 2 below, but I am not clear on this.
  2. Complex+Ligand: Simulate decoupling of the restrained ligand in the protein complex.
     * This I intend to do with SC potentials, in particular, the new Smoothstep default settings as published recently.
     * Is there special attention I need to give to the lambda scheduling given the context of my calculation, or are default settings likely reasonable?
     * Given I am decoupling a whole ligand with no shared common atoms, am I safe to use SHAKE in this context? Or is SHAKE a big no-no the moment SC is involved, requiring me to use the noshakemask setting for the ligand?
     * Are there key default .mdin toggles that need my attention in this situation?

  1. Ligand+Water: Essentially the same as Step 2, although in just water+ions and without any restraints.
     * Same generic questions as Step 2.
  2. Solve the equation relating to the Boresch restraints as they apply to my specific system.

In the context of a charged ligand, I believe I could condense steps 2 and 3 into a single step where the ligand is "recoupled" in bulk water in a single leg of the cycle, and is held a decent distance from the protein with COM restraints. Would this require any further attention in capturing the correct dG (despite the likely requirement for increased sampling given the larger overall perturbation)?

Any feedback on this process in the context of AMBER20 would be most appreciated.

Thank you kindly,

AMBER mailing list
Received on Fri Jun 18 2021 - 00:00:02 PDT
Custom Search