Performance of conformer plugin on Kirchmair dataset

Hi,





I would like to use the conformer plugin for generating input structures for rigid docking experiments.





Therefore, I wonder if you have any data regarding the performance of the conformer plugin on the well-established dataset from





J.Kirchmair, J.Chem.Inf.Model.,2006,46,1848-1861





(using different parameter settings).











Best regards, Ingo

Dear Ingo,














We have made a brief test on the given data set. (Time values are based on system clock, measured with single thread execution on a 2.4GHz Intel CPU).





The coordinate generation in single conformer mode finished in 4:15 for the 778 structures; that is an average time of 328 ms/structure.





Average run times (avg time/generated conformer) vs requested conformer counts (and failed structure count):

Code:

1162 ms/structure     5 conf.    (232 ms/conf)    0 failed 2313 ms/structure    10 conf.    (231 ms/conf)    0 failed 3876 ms/structure    14 conf.    (276 ms/conf)    0 failed 4006 ms/structure    15 conf.    (266 ms/conf)    1 aborted 5114 ms/structure    20 conf.    (256 ms/conf)    1 aborted

Regards,





Gabor

Dear Gabor,





thanks for the quick response. As I would like to use the generated conformers for rigid body docking experiments, it is of major importance, that at least one of the generated conformers is similar to a bioactive conformation (as observed in the PDB structure).





Therefore I have to deal with multiple parameters which need to be optimized (RMSD to known conformers/no. of generated conformers/diversity/run time). The ultimate goal would be to find a small set of conformers which cover a broad range of bioactive conformations with only little processing time (virtual screening is usually done using ~10^6 structures).





Can you give me any hints about the parameter settings which could be useful for this kind of requirement? What does the diversity parameter mean, diversity in terms of calculated energy or diversity in the generated conformations (RMSD)? Are there any numbers concerning the influence of the optimization level regarding runtime and quality of the results?





Greetings





Ingo

Dear Ingo,





this sounds like a good use case for our new 3D alignment plugin. The can take your bioactive conformation as a reference 3D structure and while it is kept rigid it is capable of overlaying another structure in a flexible manner. Thus ensuring maximum 3D similarity between your reference molecule (natural ligand, I suppose) and the 3D conformation of your putative ligand to be docked.





Does this sound like suitable tool for your needs? In this case you may need to generate one 3D conformation using the 3D conformation generation plugin, and the 3D flexible alignment takes care of the rest: it does explore the 3D conformational space in a continuous scale to make sure that the best overlay is found.





This new plugin will be released in version 5.2 soon.





Regards,





Miklos

Dear Miklos,














the question is not, if you can align a molecule to a given structure, but if  you can predict a conformation without knowing the exact solution in prior. Therefore I am wondering in how far one can influence the conformer generation result by tweeking its parameters towards finding a correct solution (and actually what the tool parameters mean, i.e. the diversity setting -d).














Best regards, Ingo

Dear Ingo,














Sorry for the late answer.

Quote:

the question is ... if  you can predict a conformation without knowing the exact solution in prior.

Without any other data our goal can only be the appropriate, even sampling of the conformational space. Thus we expect to have conformer near to the bioactive one among the others: currently it is not possible to bias the conformational analysis.





The referred article contained only the SMILES version of the test set without the actual 3D coordinates. If You could give us the bioactive conformers of the Kirchmair set then we might do a further comparison over our results. It could be interesting to check the RMSD of the nearest generated conformers.

Quote:

what the tool parameters mean, i.e. the diversity setting -d)

Diversity parameter defines the minimal separation of the resulting conformers. Any two generated conformer should not fit with better RMSD. This parameter does not influence the conformer generating process, this used as the filtering criteria. (However the requested conformer count refers to the filtered conformer, thus looser diversity criteria with the same conformer count results in depper analysis.)





If You have any further questions, please feel free to ask them,





Regards,





Gabor

Dear Gabor

Quote:

without any other data our goal can only be the appropriate, even sampling of the conformational space. Thus we expect to have conformer near to the bioactive one among the others: currently it is not possible to bias the conformational analysis. The referred article contained only the SMILES version of the test set without the actual 3D coordinates. If You could give us the bioactive conformers of the Kirchmair set then we might do a further comparison over our results. It could be interesting to check the RMSD of the nearest generated conformers.

I would be very interested in these results. Therefore I attached a file containing coordinates from the original PDB structures, each in mol2 format. Unfortunately some (14) of the molecules contain some unusual atoms which have no correpsonding sybyl atom types. As they are classified as Dummy atoms, the conformer generation will not produce a result for these molecules, if you take the mol2 as input for cxcalc.





Regards,


Ingo

Dear Ingo,














Thanks for the structures. We will notify You in this topic as soon any reult is produced.














Regards,





Gabor