The input files for the QligFEPv2 benchmarking experiments are all listed in the startFiles directory. Any modifications applied to the original structures obtained from the IndustryBenchmarks2024 repository (zenodo link) are described below.
- The notebook used to download the files from the IndustryBenchmarks2024 repository is available here.
- The ligand structure alignment mentioned in the manuscript is performed with the ligand alignment notebook.
- The notebook used to apply the standard Q-compatible atom namings to the pdb files and to run
qprepto create the water spheres is available here. - Finally, the creation of the perturbation networks is performed with the network creation notebook.
The JACS benchmark set is a set of 8 protein-ligand systems used to benchmark the QligFEPv2 software. The prepared ligands/structures used for our calculations are the same reported in the IndustryBenchmark2024 repository, with exception of Thrombin, which was prepared by us.
Here you can find the modifications applied to each of the targets obtained from the original repository:
All ligands and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
ILE171, SER96, SER71, PHE169, GLY291
All ligands and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
LYS89, ASP86, LEU138
N-terminal of Chain A was also minimized to remove a clash leading to infinite VDW potentials.
Ligand Changes
An additional change was introduced to the input ligand structures. We noticed poorer correlation with the experimental data when using the ligands from the IndustryBenchmarks2024 repository. Ligand 17 in the series contains a halogen meta-substituted phenyl ring, pointing towards the solvent. The protein structure 6GUK, though different from the ligand in question, displays a different rotamer pointing towards the cyclohexyl group, less solvent-exposed. The space where the chlorine group is positioned in this deposited structure has an overlap of both 2Fo-Fc
All ligands and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
GLY35, VAL40, LEU110, MET111, ALA113
All ligands and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
VAL253, MET231, LEU246, LEU290, ILE294, LEU267, MET250, VAL274, LEU235, PHE270, GLY271
No manual minimization performed. Bad clashes were only observed against water molecules, which are automatically removed before QligFEP RBFE simulations.
Prepared protein from the source repository displayed poor correlation with the experimental data. Therefore, we proceeded to use an internally prepared structure by us, generated before this study was conducted and known to work well with QligFEP RBFE calculations.
Upon checking the ligands, we noticed a need for optimization of the ligand poses. Despite the good MCS alignment for the scaffold shared by the ligands, other parts of the ligand weren't so well aligned. Therefore, we decided to perform a few additional alignments on top of the ligand preparation on IndustryBenchmarks2024 repository. The changes can be found in our ligand alignment notebook.
The protein found in the source repository contained some hydrogen positioning problems, which we attempted to fix using Maestro's Refine > H-bond-assignment tool. Further, some amino acids were placed in the sequence in the incorrect order. Those were fixed by manually reordering them.
The resulting structure, however, resulted in crashes during the FEP, which wasn't observed for any of the other targets used in this study. Therefore, we proceeded to use an internally prepared structures by us.
All ligands and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
LEU903, TYR980, GLY984, PRO982
All ligands and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
val27, gly28, tyr32, lys52, ile79, his102, asp103, asn156, leu158, arg356
All ligands and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
ile1084, gly1085, met1160, lys1161
All ligands and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
arg119, pro121, leu160, gly217, ala218
met289, his293, cys339
Ligands 20, 41, and 42 and respective protein structure were loaded in Maestro. A minimization step was applied to the following residues by manually selecting them and minimizing with the Ctrl + m command:
val214
Following that, ligands 44, 47, 52, 53 were loaded. A second minimization step was applied to the folliwng residues by manually selecting them and minimizing with the Ctrl + m command:
leu238, ile241, his242
The following water molecules were removed: 905, 914, 944, 993, 998, 1022, as they were clashing with other HOH O atoms and not the closest to the protein residues.
Removed the atoms:
ATOM 1 CH3 ACE A 0 87.427 98.432 260.536 1.00 0.00 C
ATOM 2 C ACE A 0 86.302 98.808 261.499 1.00 0.00 C
ATOM 3 O ACE A 0 85.472 97.963 261.827 1.00 0.00 O
ATOM 4 1H ACE A 0 87.325 97.362 260.246 1.00 0.00 H
ATOM 5 2H ACE A 0 87.370 99.072 259.627 1.00 0.00 H
ATOM 6 3H ACE A 0 88.411 98.593 261.030 1.00 0.00 H
...
ATOM 284 N NME A 16A 79.441 97.537 254.883 1.00 0.00 N
ATOM 285 CA NME A 16A 80.445 98.441 254.341 1.00 0.00 C
ATOM 286 H NME A 16A 78.820 97.041 254.262 1.00 0.00 H
ATOM 287 1HA NME A 16A 81.022 98.883 255.184 1.00 0.00 H
ATOM 288 2HA NME A 16A 81.132 97.880 253.668 1.00 0.00 H
ATOM 289 3HA NME A 16A 79.949 99.249 253.759 1.00 0.00 H Ligand Changes
An additional change was introduced to the input ligand structures. We noticed poorer correlation with the experimental data for the edges including ligand 43 from the IndustryBenchmarks2024 repository. This ligand in the series contains a halogen (Br) meta-substituted phenyl ring, pointing towards the solvent. The protein structure 6HVI with the co-crystalized ligand 38 of the congeneric series also contains a meta-substitution of the phenyl ring, but on the opposite orientation than ligand 43. Therefore, we decided to flip the cyclohexyl group in ligand 43 to match the observed orientation in the protein structure. Doing so, we observed a better correlation between the calculated and experimental data, supporting the decision to use this pose.
phe113, his114, thr219, glu249, asp489, lys492
The following residues were minimized to better accommodate the ligands in the binding site:
glu376, leu377, gly378, val385, asn457, asp512, phe513, lys402, gly454, ser379, lys375, phe382, lys458
Further, other amino acids were minimized to avoid protein-protein clashes.
Finally, the orientation of the protein's hydrogen atoms were refined using Maestro's Refine > H-bond-assignment tool by checking the boxes:
- Sample water orientations
- Use PROPKA pH: 7.0
No manual minimization performed.