In many cases decoys for validation of virtual screening algorithms are molecules derived from public databases (and so, they are synthetically feasible). However, the generation of virtual decoys (i.e. molecules that may be not synthetically feasible) can be desirable. can someone suggest strategies / software?
One way to do this, is to use the DUD (a directory of useful decoys) library for benchmarking virtual screening. DUD is designed to help test docking algorithms by providing challenging decoys. It contains 2,950 active compounds against a total of 40 targets, and for each active, 36 "decoys" with similar physical properties (e.g. molecular weight, calculated logP) but dissimilar topology.
When I was generating decoys, I came across this paper: http://www.ncbi.nlm.nih.gov/pubmed/22539671. You can download their software or, if you're interested in programming, it is relatively simple to write your own decoy generator using Python and Open Babel.
It's true! But DUD holds molecules of ZINC database. I mean the computer-driven generation of "virtual" decoys intended as molecules not necessarily available and/or synthetically feasible.
If the data on your given target are available, one alternative is to use the experimental results of the PubChem Bioassays as long as you can curate the compounds and restrict your selection to confirmatory results with at least IC50 for them.
With DecoyFinder (http://www.ncbi.nlm.nih.gov/pubmed/22539671, as Malgorzata poited out) by default you find decoys from molecules in the ZINC database, but you can use any SDF or mol2 files as the source of decoy "candidates". You could take a database of virtual molecules and use it as the source of decoys for Decoyfinder, and all the decotys it will find will come from those virtual molecules, thus being "virtual decoys".
Paul,
Virtual decoys in some cases may be less biased than synthetically feasible molecules with respect to physical similarity. In addition they can be made on any molecular scaffold.
Moreover, if the high rate of false is usual in the virtual screening, the story is (or at least, should be) quite different in the molecular modelling.
For example: it is under development the modelling of the effect of specific modifications on a certain molecular scaffold and it is known that the glucuronidation may inactivate a particular class of molecules. The most commonly used benchmarks for decoys do not contain, as the best of my knowledge, glucuronidated molecules. In this case, the generation of glucuronides and analogues (not necessarily synthetically feasible) from a particular set of molecules may be useful.
If the goal is to test whether a docking algorithm works, a DUD set would be best as they are widely accepted benchmarks. However, if you're trying to ascertain if your protocol for docking is capable of discerning active from inactive (decoy) molecules, then you want inactives with similar physiochemical properties to your active molecules.
If you have data on molecules which are inactive against your target, I'd suggest using those. Alternatively, Decoyfinder (which has already been suggested) or a manual development of a decoy set using the properties of your active molecules would be acceptable.
- Badges
- Science topic
- Similar topics
- Filing