Correlation Between Predicted And Experimental Binding Affinities In Docking, Good Examples?

Question

[Flow]

I am new to the field of docking. I am learning Autodock, and I can see that although it can sometimes predict docking poses in redocking rater well, fails very frequently when trying to correlate scoring function for the predicted pose with experimental binding affinities, for a set of compounds. At the moment I am trying to find publications where some docking program/procedure can obtain a good correlation in affinity prediction but I do not find anything impressive or just acceptable, do you know about any good result in this context with the reference?

Answer 1

[Jan Kosinski]

I think that what you get is the current state-of-the art: it's much easier to predict the correct pose than correctly rank ligands by affinity. Limiting yourself to a single docking program is also not a good idea, as according to several studies, there is no single program that performs well on all docking targets.

You may look at this study: Can we trust docking results? Evaluation of seven commonly used programs on PDBbind database. (But also read comments on this article on annotatr)

Somewhat older paper: A critical assessment of docking programs and scoring functions.

[Flow]

thanks for the answer. nevertheless, I am sure some good papers exist around, although I can not find them at the moment

[Jan Kosinski]

Even if you find such a paper, does it tell you anything? If let's say you find that one paper out of 100 papers about docking small ligands got good correlation for a particular protein and for a particular set of ligands, wouldn't you expect that simply by chance? In other words, given so many groups trying to get this correlation for their own case studies, at some point someone must get good correlation between theoretical and experimental values simply by chance.

[Flow]

i agree, but curiosity rules science, and I am being just curious and want to know, is it ok?

[Jan Kosinski]

Is it ok, and I also would like to have a look on such examples. What I just want to say that one muse be careful with using such studies as examples of successful binding affinity predictions.