Hello,
I recently started working on QSAR classification modeling and used some of the already existing protocols in PP (many thanxs to all the contributors). Some of the techniques using FCFP_6 on SVM, Bayesian etc. works really well on my example test cases.
As next step, I am trying to use LDA, BLR etc., purely based on preselected physicochemical descriptors, in the hope that i can have a final equation defining the role and contribution of each of these descriptors.
Many such papers, such as by Mark Cronin's group, exist. In one they explained making such equation (
p = -0.19 ClogP +0.03 SHBint2 -0.1 SHBint3_acnt +0.33 SHBint6_acnt ... +1.856) using LDA and another by using BLR method in SPSS software using dataset of (without any activity value) antibiotics and non-antibiotic compounds (
http://www.ncbi.nlm.nih.gov/pubmed/12132888). How would you proceed in doing something similar in PP?
Also ~2 years back Tropsha et al published a well refined strategy (workflow) to make QSAR models, validate them and use them in various experiments etc. (
http://onlinelibrary.wiley.com/doi/10.1002/minf.201000061/full). Its a well known workflow and lots of papers have been published following it. I was wondering if someone has already 'worked/implemented in PP' on it.
Any response/comment would be greatly appreciated.
Thank you
nar
.