Let's talk
Contact us

Q&A: Pharmacophore modeling for potential drug discovery

Dr. Apurba K Bhattacharjee, Georgetown University

Q: What are the current approaches to screen for compounds that would fit into a given pharmacophore?
A: Pharmacophore based searching, structure similarity based searching, docking at the active site, dynamic docking and use of machine learning (AI) tools. 

Q: How to choose the 3D pharmacophore model software programs considering parameters either molecular filed or substitution pattern or a combination of both? 
A: There are various functions available in different software. You are to select them based on your specific needs. Discovery Studio (BIOVIA), COMFA (Comparative molecular field analysis (CoMFA) on the training set of active compounds has the predictive ability to test on additional compounds. There are a variety of them available commercially.

Q: How to do a 3D model when no active ligand(s) available for a known target?
A: Some experimental information of a ligand or target active site details should be available. You may hypothetically design a pharmacophore model based on complementary information from the active site of the known target. Use the hypothetical pharmacophore model as template for search of compound databases may enable you to find ligands for the target. You will even be able to identify a few compounds from the search and make a 3D model.

Q: How many compounds does one need for an effective “training set?”
A: If you use statistical methods like Catalyst (Discovery Studio) methods, you should ideally require at least 15 compounds in the training set. If you start with QM (quantum mechanical) approach to develop molecular electrostatic potential profiles, you can perform the task with lesser number known active compounds.

Q: Is it possible to provide complete molecular flexibility to the target protein?
A: Yes. Dynamic docking can be performed. You may also consider water molecules at the active site of the target.

Q: How to address a major limitation of Ligand-based pharmacophore dependence on pre-computed databases that contain a limited number of low-energy conformations per molecule?
A: Most of the procedures for generation of compound databases useful for pharmacophore based (not structure of ligand) search recommends multi-conformer generation of molecules within 0.0-20.0 kcal/mol. Catalyst, Discovery Studio (Biovia) etc already have procedures for such database generation. 

Q: In your examples you have illustrated the use of Pharmacophore in already known and validated receptors. In case if the receptors are not known or validated how does Bioinformactis/AI assist us to discover complete novel receptors and pharmacophore. 
A: Not really. Pharmacophore based procedures are most useful when the receptor structures are not known. Only a handful of active disease specific compounds (15-20) should be enough. Even less number of active compounds are also possible but will have to use QM methods for generation of molecular electrostatic potential (MEP) profiles. The MEP can be an useful guidance for pharmacophore generation even for small training set of active molecules. Approach for Bioinformatics/AI tools are useful only when you have large (really large) number active and inactive compounds in a database. 

Q: From the standpoint of pharmacophores, in your opinion, which is the least and the most favorable rule of the Linpinsky’s rules?
A: From Druglikeness point of view, Lipinsky’s rule is empirical. However, it is very useful when you have a short list of 5-10 lead compounds to promote, the rule is extremely helpful. 

Q: Specificity or Sensitivity? Which one will you prefer for pharmacophore modeling validation?
A: The goal will undoubtedly be specificity.

Q: For unknown target, what would be the average number of molecules needed to build a good pharmacophore model ? Which parameters (e.g. IC50) would be mandatory to build this model?
A: A training set of 15-20 (10 highly potent, 5 intermediately potent, and 5 inactive) molecules should be ideal training set. Any of the experimental data from IC-50, ED-50, LD-50 etc may be used to build a pharmacophore model. 

Q: Which software do you recommend for in silico ADMET property predictions?
A: I cannot recommend any single software. There are many good commercial software available now a days. Even free software are available for downloads. I have used TOPKAT from Discovery Studio many years ago.

Q: Can you please provide a list of general references about the methodology

Q: How to find out trend/correlation between biological activity and physicochemical (ADME-Tox) properties of the molecules using 3D modeling?
A: Again you would require experimental data (or reliable in silico ADME-Tox) for a set of at least 15 compounds to create a training set to find out the correlation and trend .

Q: How do you anticipate/address active or potentially genotoxic metabolites?
A: You need to know the metabolites. In order to know the metabolites, you will have to have the knowledge of mechanism of action of the active compounds. Please remember when pharmacophores are generated, metabolites are not considered

Q: How often (percentage wise) does an in silico ‘hit’ translate to biological activity?
A: If iterative processes of refining the pharmacophore, search for databases and in vitro testing are carried out, the chance of a very reliable pharmacophore model could be achieved. 

Q: Given that good antitumor drugs are usually not having a good water solubility, one question here is that through this computer drug discovery technology, we may rule out some good candidates that have good antitumor activity for example but due to its poor water solubility (but this can be improved by formulation), such compounds may be ruled out without further consideration.
A: Yes, you are very correct. That is why a computational chemist should be always in consultation with a medicinal chemist before promoting a lead compound.

Q: Where this database come from?
A: There are many commercial compound databases, such as Maybridge, ChemNavigator, Zinc etc. I have used our own in-house database (WRAIR) of about three hundred thousand compounds.

Q: Can we use electrostatic surface potential predicted from QM method to study ligand-protein interactions? If yes how interpret the results?
A: Absolutely. Molecular electrostatic potential (MEP) profile is the interaction pharmacophore of a compound. You can even calculate MEP in water or other solvents using the dielectric values of solvents.

Q: How people get exposure to Nerve agents these days?
A: Mostly from pesticides and certain kind of insect repellents commonly used crop protections.

Q: Are there any studies on how often in silico screening finds real hits? I have seen a few examples where it worked, but I have seen many more that haven’t. Gives hits that are not active when tested?
A: You see experimental testing is absolutely necessary. Many cases are there in literature where it failed. However, iterative refining of models with testing usually produce good results. Remember in silico approach is another tool only like x-ray crystallography, nmr, spectroscopy etc. 

Q: If a compound has a clogP of 6 or 7, but satisfies all other criteria, should it be excluded right away?
A: It should be tried for in vitro testing before discarding it. Lipinsky’s rule is an empirical rule only not hundred percent guaranteed.