i am working on MOE for Structure base pharmacophore model based virtual screening. The target, i am working on, have the 2-4 known active compounds. I have used to 2 pdb complexes, with 2 of those known ligands, for the development of SB pharmacophore. So the question is should i go for validating my pharmacophore? is it necessary in my case study? if yes, how should i validate my pharmacophore model when i have only 2-5 active known ligands? the ratio of active and decoys?
Hello Suresh kumar,
I have used the co-crystallized ligands (i.e. inhibitors) of proteins to develop the shared pharmacophore using fingerprints of protein ligand interactions in MOE. I want to use this pharmacophore for virtual screening. However, before this, i want to validate the pharmacophore to standardize its prediction capacity. But, unfortunately, the target has 2 to 4 known inhibitors so far. So what strategy should i adopt to validate my pharmacophore model.
I have seen that descriptor based data mininig, as you mentioned in your comment, to search the similar active compounds is mostly used to supplement the test database with decoys during validation, to avoid the possible biasness in validation. I wonder, how it would be supplementing the active compounds of test database.
Please also guide about the ratio of active vs decoys in test database. Thank you very much for your time!
As I understand it, you've derived the pharmacophore from two known actives co-crystallised with your protein structure. Are the other 4-5 known actives which you plan to use for validation different from these two? Ideally you should have more known actives for validation, but since you don't I suggest at least go ahead with the molecules you have for validation. Validation with known actives is absolutely necessary for a pharmacophore hypothesis, to see if it can distinguish the actives from random molecules or not. W.r.t ratio of actives to decoys, there are no specific rules as such but most studies usually use about 1000 decoys and ~100-300 actives if available. Some prefer a ratio of 1:1000 (actives:inactives). I can suggest that you first try to validate with a data set of the 5 actives and about 1000 decoys, if you can get good enrichment (unlikely) then your validation will be very good. If not, then you can reduce the number of decoys subsequently - 500, 200, 100 ..... but try and maintain at least 1:10 ratio. And make sure the 2 ligands already used for pharmacophore generation (training set) are not a part of your test set. Also do not include any molecule for which experimental inhibitory activity is not known against your target as actives for valdiation - i.e. by similarity searching of databases with known actives. A molecule cannot be considered as an active just based on similarity.
- Badges
- Science topic