Re: CCL:Docking software vithout validation

Dear All,
 I know very well that validating a docking software is rather time
 consuming and needs some experience. Note I spent some time (years) in a
 group with validating and developing a docking software that is used
 in-house at a durg-discovery company.
 Docking softwares originally were developed to restore Xtal ligand
 structures quickly and virtual screening were not always in mind at the
 birth of these tools. Luckily, they can be used for hit-id as well, that
 is a very handy "by-product". Though validation is expensive but
 necessary: as it is unacceptable from a car-manufacturer to leave its
 products validation for the customers, it is true for all other products -
 for software as well. Choice is a result of comparison: with validation
 data I can choose between A_dock and B_dock.  I do not want to impose a
 compulsory validation set on any developer but sticking to a cleared set
 that is used by other programs also makes the choice easier for the
 customer. The source of these programs likely will not be freely available
 for quite a while and it is understandable. Docking is still a new
 technology and the competition is already strong.
 As for me the main validation points would be (and it is about the
 scheduling the validation also):
 i)    reproducing RMSD on a large target set
 ii)   checking selectivity (ie with cross-docking matrix)
 iii)  proving docking can select active compounds from a random library
 iv)   providing tools for post-processing (discarding false positives,
       opportunistic binders)
 One can argue the last step is not the docking software's task but I want
 to emphasize my point of view is that  such a "docking platform" has
 a role to select putative leads from a large virtual library. As for this
 step iv) is required because the crude run-results are heavily polluted
 with false positives.
 Usually people are expecting docking to reproduce not only the exact X-ray
 conformation of the ligand but the free energy (dG) values as well. My
 opinion is that this is not the task of docking: the dG values in the
 literature are quite unreliable and there are other problems with
 the scoring functions also.
 PS: sorry not to include all the mails I responded to, but this lengthy
 email would be even longer.