Please refer to the following research article (its freely available on ACS)

http://pubs.acs.org/doi/abs/10.1021/jm201371y

the biological activity measured on cell lines include both target binding + disposition of the compounds (non-specific binding, metabolism, membrane binding, etc). If these later effects are same for all compounds, then CoMFA model might be useful.

since you have good number of compounds (~80), I would strongly recommend to exclude the compounds with subquantitative values. You may probably want to use them as test set compounds ?.

Hi there.

Prerequisite for QSAR are here ok in my opinion. If you want to perform 3D-QSAR you have 2 options. Methods that require 3D superposition of molecules and approaches that mine 3D-fields and producing vectors. CoMFA belongs to the former. Docking solution can be a good trick to align molecules is a sensible way. Pharmacopore can be another interesting way of alignment (in Sybyl environment GASP is very handy). But conceptually 3D-QSAR has little to do with binding modes or bioactive confirmations.

Lots of free and open-source alternatives do exist. Please have a look at www.click2drug.org for a well maintained list of modeling/compchem/cheminformatic tools.

Enjoy!

Antoine.

Pleasr see my paper "Prediction of........." It will be of some help to you.. Multitarget Virtual screening exercises can definitely save your time and efforts.

Since you have similar scaffolds in the dataset it will be easy for you to align for CoMFA analysis. I suggest you to check the conformation of the most active compound in the dataset. Is this compound bound to any cancer causing target?? Then check in PDB its binding conformation etc. If not, then try to perform ligand based alignment approach. Compound whose activity is not exactly known can be taken as test set or external dataset for vaildation of your CoMFa model. Hope this helps.

Thank you so much for all your suggestions!!

One of the main problems with traditional comfa in my case was that I had seen a large variability in the r2 and q2 values depending on the what I chose as the starting conformation for alignment. Using just the energy minimized values did not work well (r2 ~ 0.6-0.7, q2 ~ 0.2-0.3 with large errors), and while docking seemed to improve the values (r2 ~ 0.8, q2 ~0.5) it can get better. Further, I need to use at least some of the inactive compounds (compounds with a '> x' value) to ensure that I have a more uniform spread across the bioactivity range; removing them really messes up the test set results.

Using topomer comfa seems to be generating better q2 values till now (~0.6 - 0.7) quite simply because it ignores the starting conformations and focuses on where the variability actually is on the scaffolds.

@ Dr. Natesan - thank you for pointing out your publication. I will reference it when this work gets published.

@ Dr. Sharma - I do not have fulltext access to your publication. Would you be able to send me a pdf copy of the same?

@ Gopi - I do not have structural information for the most active compound. Hence was trying to align the compounds on docked poses which can best explain the observed SAR. Further, some of the targets have not been crystallized. Wouldn't homology models and docking into them compound the errors?

@ Dr. Daina - thank you for the pharmacophore suggestion.