1 of 5

Predict among a series of lead candidates, those that will bind more strongly to the therapeutic target
Calculate free energy of binding differences between a parent candidate drug and its analogues
Automate free energy perturbation calculations
Provide an easily accessible web-based tool to facilitate the set-up procedure for simulations

ΔΔG?

Automating Free Energy Perturbation (FEP) calculations

FEPrepare

2 of 5

ΔΔG(A->B) = ΔG₂ – ΔG₁ = ΔG_B - ΔG_A

If ΔΔG(B->A) < 0, B is favored for binding

Free Energy Perturbation (FEP) calculations

Sampling with Molecular Dynamics

needs thousands of CPU or GPU hours

FEP methodology

Zwanzig formula in complex and in solvent:

Absolute need of HPC resources

Ζwanzig, J. Chem. Phys, 1955

Cournia et al., J Chem Inf Model, 2017

MBAR used in practice

3 of 5

Free Energy Perturbation (FEP) calculations

FEPrepare workflow

FEP/MD calculations in NAMD using

thousands of CPU or GPU hours

OPLS-AA

parameters

Resources for 1 calculation:

4,480 CPU hours

(NAMD 2.14)

600 GPU hours

(AMBER 20)

Resources for 100 calculations: 448,000 CPU hours

(NAMD 2.14)

60,000 GPU hours

(AMBER 20)

To avoid any inconsistencies in the atom naming that might occur and could cause NAMD to terminate abnormally, a Python script, which renames all atoms of the reference and perturbed ligands according to their coordinates

In the NAMD dual-topology scheme, both the reference and perturbed ligands exist in the same topology file (hybrid RTF file). In order to

avoid perturbing all atoms of the reference ligand to the target ligand, an algorithm needs to be applied in order to reduce the required perturbations of atoms during the transformation and to speed up the convergence of the algorithm. Using our scheme, the common moiety between the reference and perturbed ligands retains the same parameters during the simulation, and only the perturbed area is perturbed.

The atoms with the new charges will be included in the hybrid RTF with different names from their reference counterparts, in order for NAMD to recognise them as different atoms.The produced hybrid RTF will include the common part of the two ligands, the perturbation area before the transformation, and the perturbation area after the transformation. So, essentially, it will include the whole reference ligand and, from the perturbed ligand, only the perturbed atoms that were affected by the partial charge distribution procedure.

In addition to a hybrid RTF, a hybrid PDB file needs to be produced, accordingly. Because the perturbed atoms that had their charges changed had also their names changed, with respect to their reference counterparts, the PRM file of the perturbed ligand needs to be

updated, to account for the new names.

For the complex leg of the simulation, it is necessary to have a protein-ligand complex PDB file. To build the complex PDB, the protein PDB and the hybrid PDB files need to be

combined, simply by concatenating the two files and fixing the atom numbering of the ligand at the end of the new file.

4 of 5

Free Energy Perturbation (FEP) calculations

Results

CK666

AI003

ΔΔGs are the same between setting-up the simulation manually and after using the webserver

Perturbation	ΔΔG_NAMD	ΔΔG_FEP+
CK666-AI003	-4.07 ± 0.06	-4.50 ± 0.41

kcal/mol

Perturbation	ΔΔG_NAMD
AI003-CK666	3.61 ± 0.06
CK666-AI003	-4.07 ± 0.06

Forward-backward calculation

Resources for 1 calculation: 4,480 CPU hours

100 calculations: 448,000 CPU hours

Zavitsanou et al, J Chem Inf Model 2021

Preparatory access:

ARIS supercomputer (GRNET)

Project acronym:

FEPrepare

5 of 5

FEPrepare: Web server implementation

Zavitsanou et al, J Chem Inf Model 2021

Contact support: zavitsanoustamatia@gmail.com, mpapadour@bioacademy.gr

http://feprepare.vi-seem.eu/