[Dock-fans] RE: [Dockdev] Docking and Scoring algorithms
Vincent.Leroux at edam.uhp-nancy.fr
Sat Feb 26 11:26:39 PST 2005
I agree that most scoring functions are not correlated enough to the binding energies. But for me this is not as bad as it seems.
Docking -even with the best scoring functions- will never be precise enough "alone" if one starts from millions of molecules. But some methods can be helpful.
For example, QSAR. You can choose to filter out a molecule only if docking is bad for it AND when QSAR also says "nothing special" regarding the target. But at the risk of ending with 100k candidates instead of 30k (30k is still too much anyway...)
You can use several methods of increasing precision, but testing some "bad" molecules from time to time just to "validate" the previous calculation. It does not matter if you have false positives at one point (if one method says you don't, that means the lost your time using it...). But false negatives are a problem, so from time to time, you can check for that (like for 1% of the rejected molecules). If you detect a significant number of false negatives, then the whole calculation has to be stopped. Anyway SOME false negatives are not that terrible, if you find enough new binders amongst the final hits...
A program like SILVER can help you if you want random picks with enough diversity from docking results
(those from GOLD, for instance). You can also classify the "hits", but qualitatively, and you need "markers", for example the existence of certain H-bonds that are expected (from experimental data). I agree this is not enough.
For sure I would be happy if one single scoring function would be enough to predict binding energies with docking methods, without being too CPU-costly. But well for the moment you cannot have BOTH precision and speed... You want speed? You have docking. You want precision? You have MD and FEP...
I only see docking as a qualitative method : good to give clues, good to filter large molecular databases, and very valuable when one does not know "where to start" with a new target. But if I want to have more precise information, I would use MD. Several times docking calculations gave me excellent ideas in order to mutate some binders that I would not have had myself. Good enough for me...
I think docking will never be able to find every single hit for a target amongst millions of molecules. Docking is unprecise by nature. One reason is that it is not meant to model solvent effects. Of course you can tweak the scoring function, but you will never be able to take care of all types of solvent effects that can be observed in a MD simulation, just because many of them seem to be time-dependant. For docking, the way the target is modeled (with included water molecules - which? where? or not) can change dramatically the results in some cases... If you are unsure of your receptor model (and if you think that the presence of some water molecules can also depend on the ligand - then you are), what is the need of a more precise scoring function? Of course, at the same speed, a better scoring function won't hurt. But it won't be enough also...
Hope that does not sound too pessimistic...
About complaining chemists : if they really ask for FEP-like calculations as fast as docking, time for them to return to Earth... And if computer simulations were faster and more reliable than experiments, wouldn't many of them lose their job? I am not sure that is what they need ;-)
********************** A 24/02/2005 20:33, Szilveszter Juhos a écrit:
I do not want to repeat issues written in the previous mails, simply
want to emphasize that there is a real need for fast and accurate
binding energy estimation. Fast means hours instead of weeks, accurate
means being good enough to rank hits. When there is a huge virtual
library with millions of compounds and finally one gets 30K hits, it
is not so obvious to filter out the bad ones.
Using FEP one have the trivial problem of loosing diversity. If I want
unseen chemical entities how to get their dG without measuring it in
the wet lab? Not speaking about the fact that many of the dG values
form experiments are rubbish, simply can not be used for docking
validation (or for anything at all) (sorry).
To cut it short, what I want to say is that chemists always will
complain that docking is not providing correct binding energies. It is
a convenient excuse to say "docking is not for dG estimation" but they
are complaing righfully since they badly need a (fast, accurate)
method to do it. Probably it will be something but docking or FEP but
since that "if all you have is a hammer everything looks like a nail".
Dock-fans mailing list
Dock-fans at docking.org
Vincent LEROUX PhD student
Equipe de Dynamique des Assemblages Membranaires,
UMR-CNRS 7565, Université Henri Poincaré, Nancy I,
BP 239, 54506 Vandoeuvre-les-Nancy, France
e-mail: Vincent.Leroux at edam.uhp-nancy.fr
More information about the Dock-fans