1. What is the criteria for the selection of training and test set for QSAR model?
2. Is it compulsory for 3D-Model, you should choose best 2D-QSAR models training and test set data without changes? If answer is "YES" then what is the importance of 2D-QSAR model? and, If, answer is "NO" then some times best 2D-model (r2,q2, pred_r2 aprox same and > 0.7) is not gives good result in 3D-QSAR. Why?
3. How to apply validation techniques like Leave-one-out, Leave-many-out and Y-randomization on a QSAR modal?
4. Is there any fix range for acceptable residual error, r2_se, q2_se, and pred_r2_se?
5. Please, can you suggest good book or article on QSAR and drug designing?
The training set should molecules with a wide range of dependent variable. For example, if your dependent variable is IC50, then the training molecules should have compounds with low IC50 (good activity) and high IC50 (poor activity). Also, compounds with different kinds of substituents should be included. Like, electron donating groups, electron withdrawing groups, bulky groups etc.
Not sure of 2nd and 4th questions.
For the 3rd question, I think it depends on the software you are using. I used open3dqsar (http://open3dqsar.sourceforge.net/). In that LOO, LMO are quite easy. It's done through a simple command.
The rationale on whether one needs a 2D or a 3D QSAR model is easy to formulate in general terms, but difficult to apply in a specific situation. Anyway, in general, if you "feel" that a given biological activity can be explained by by physical properties of chemical compounds (or their fragments), which are invariant with respect to compounds' spatial orientations, then a 2D QSAR would work best. Sometines, topological descriptors can be used as proxies of physical properties. However, if your data set "tells" you that the biological activity varies due to particular spatial arrangements of functional groups (aka. pharmacophores), then you'd need to try a 3D model, such as ComFA.
For your 4th question, i suggest you reading some of these authors:
http://onlinelibrary.wiley.com/doi/10.1002/cem.1331/abstract
http://onlinelibrary.wiley.com/doi/10.1002/qsar.200610151/abstract
http://www.scielo.br/scielo.php?script=sci_arttext&pid=S0103-50532009000400021&lng=en&nrm=iso&tlng=en
I believe that you could find the answers to the majority of your questions in the software QSARINS and also reading the included references of papers and book chapters. QSARINS is freely available for researchers asking a licence in my web: www.qsar.it
- Badges
- Science topic