Docking-I_sc values do not correlate

There are a few factors to consider…

General:

• Dissociation constant is different from binding energy. They correlate and some paper even plot kcat/KM vs. ∆∆G: https://www.nature.com/articles/ncomms12965 But the former depends on alternative bindings etc. and a form of steered MD (molecular undocking) is needed in some cases to better hone in on the dissociation constant, especially for proteins with narrow entranceways. In this Stack Exchange Bioinfo Q&A for example I discuss a nasty protein, Cytochrome P450, whose enconter complex ∆∆G has no influence on the specificity/reactivity, which is fully dictated by how the ligand gets in. I once heard for a protein and its substrate being described as a “ship in a bottle”.
• Accuracy. The accuracy of any free energy scores is 1 kcal/mol or worse for any computational biochemistry method, even FEP by MD.
• A priori ligand properties. The ligand itself without protein influences how it well it binds: https://academic.oup.com/bioinformatics/article-abstract/36/3/758/5554651?redirectedFrom=fulltext
• Partition coefficient. A variant of the above: the ∆∆G is the difference from unbound, which is in aqueous solvent. What if one has a much higher partition coefficient? How this influences Kd assays is causing weird buffer specific results —There is no simple way to calculate the ∆G in membrane of a ligand with Rosetta.

Specific:

• Alternate protonations of sidechains are not sampled, except histidine tautomerisation.
• Implicit solvent. Rosetta uses implicit solvent. Solvent interactions may help with the binding. There is a solvent placing protocol called SPaDES/hydrate.
• Stuck in a local minimum. Maybe the conformation of the ligand and sidechains and backbone endowning the lowest minimum was not sampled —how does the distribution of ∆∆G from the replicates look?
• AtomTypes may not be ideal fit—nitrile is wrong with `mol_to_params.py` for example.
• How were the Gasteiger-Marsili partial charges calculated? If mol2 file from OBabel, then from there.
• Rosetta does not have a Drude oscillator for polarisable ligands. If the ligand is strongly polarisable or tautomerises, parameterising both may help.