Sounds like an interesting question, but probably needs some reformulation. It is unclear why differences in docking poses and scores for different ligands to the same protein need to be investigated. Isn't it the very purpose of docking-based screening to obtain different scores for different ligands and use them to rank the ligands?

Dear Dimitri,

thanks for your follow up.

I am performing analysis where I databse is for example:

1. p53 X-ray holo structure 1

2. p53 X-ray holo structure 2

each holo structure have different ligands, same sequence, totally identical structures apart from ligands.

I have the way of analysis that I am performing that tells me which structure manages to identify more exp confirmed binders as the top scoring compounds among compounds that are not binding to this pocket.

Now when I observe differences I am trying to think of analysis to perform on binding pockets to see why one structure manages to identify "true ligands" as higher scoring one when compared to the other?

I can think of calculating number of H-bonds, binding pocket volume, maybe something related to sequence etc. But would like to first think of all analysis that makes sense to perform in this cases.

Best,

Srdjan

Different complex structures may show subtle conformational differences in the binding pocket that can have a major effect on the docking scores in rigid body docking. Newer algorithms try to compensate for this by allowing for flexibility on the side of the ligand, and, in some cases, on the side of the receptor. Alternatively, instead of docking to a single conformation of the receptor, an ensemble of receptor conformation is generated by MD or rosetta backrub as an input into the docking process. All this of course significantly increases the computational load.

If starting from different complex structures yields very different docking solutions, the algorithm does not allow for sufficient receptor flexibility to compensate for the differences in the starting structures.

Normally, you do not directly compare the top solution, but first use a clustering algorithm to see whether there are different populations of solutions. You may then perform additional energy minimization on all solutions above a given threshold score, and see whether re-scoring of these optimized solutions gives you a better match to experimental ranking.

I am sorry to come up with more questions. What does the statement "totally identical structures apart from ligands" mean? Does this mean that RMSD between these two protein structures (without ligand) is 0? If so, then magic can be the only explanation why do they behave differently with respect to a docking algorithm.

In case the structures are not "totally identical", then it is perfectly normal that one of them does a better job in a virtual screening study, but it is highly unlikely that you would ever be able to find out why. Calculation of binding free energy for a ligand-protein complex is yet an unsolved problem. Even imperfect theoretically thorough solutions require weeks of computational simulations. That is why any attempts to explain free energy differences by correlating them with simple properties, such as, numbers of H-bonds, volumes, etc, may only have limited success.

Just use the structure that does a better job for your production screen (although there is no guarantee that this structure will be "friendlier" for the screening library compounds too).

Protein flexibility can give rise to difference in binding affinity and binding interaction of a drug with its target protein.

Multiple receptor conformers based molecular docking was performed through computational chemistry in the attache literature.

Best wishes