Finding new molecules with interesting biology using ligand information only:

1D methods like property matching.

2D or graph-based methods (fingerprints, substructure searching).

3D methods ( shape matching, electrostatic potential matching).

You can do all this with tools from OpenEye, which are free to qualified academics.

Discovering new drugs on the other hand is a much more difficult problem...

Similarity search?

Shape similarity - can be used for virtual screening. Eg. ROCS by Openeye software.

there are many different favors of pharmacophores, like feature trees and topomer search which use a kind of 2.5 dimensional searching, then shape similarity using hard or soft (like Rocs) shape definition, feature pharmacophores also combined with shape like accelrys catalyst or unity.

There are pharmacophore like fingerprints. Often substructure searches or fingerprint similarities behve quite well(there were gSK papers in2006 or so).

then there are pseudoreceptor concepts like from Ajay, in QXP or in the accelrys software.

There is comfa in many flavors like e.g. topomer comfa.

But all of these methods lack the same problem. All the hits will always reflect just the information from the starting ligands. You will not be able to find the new class with different binding mode addressing a new sub pocket. Or if you will, your method sucks.

Well, to be honest everything mentioned thus far (QSAR, pharmacophore perception, molecular similarity) is related to structure-activity relationships, so depending on how generous you are with your definition of "QSAR" you might put nearly everything related to ligand-based design into that one category.

One body of knowledge that you can mine is prior screening data for assays that are based on targets homologous to your own. This nudges you away from SAR and toward something closer to activity-activity relationships.

That aside, however, I don't you will find many ways to grapple with ligand-based design that do not rely on probing the direct or indirect effects of ligand structure.

Screening molecule data bases using docking algorithms (DOCK, GOLD, etc)

Fredy Sussman

Ligand-based virtual screening of small molecule databases based on 2D (and/or 3D- shape analysis) similarity search.

Pharmacophore based virtual screening on basis of features selected during Pharmacophore modeling can be used for ligand based designing.

if you have PDB of the target then you can go with multiple docking using autovina and those show good binding you can re doc it using Autodock if you need any help the do let me know in this regard

Fragment based virtual screening

Database search approach

Based on the targets, receptor mapping

Shape similarity - can be used for virtual screening by Openeye software - ROCS , BROOD,POSIT and SZYBKI.

Thank you very much for posting.

Yes i agree for shape based screening. But it will help you when you want to screen against specific potent/drug like molecule (specifically one which just entered in clinical trial). But what will be the case if you have 60-70 molecule each shows more than 90% inhibitory activity. In this case we cant go for shape similarity for each molecule.

@ Andreas Göller,

Yes i gone through topomer confa, but i think it sounds like QSAR and shape similarity (especially topologically) in combination.

I got the article of pseudo receptor by ajay but facing a problem while retrieving gsk 2006 article, can you provide its reference.

@Gerald Lushington.

Dear gerald you mention that there are many ways to grapple with ligand-based design that do not rely on probing the direct or indirect effects of ligand structure. Can you mention that ways.

Actually what i observe, when we specifically talk about ligand based drug discovery, when you have no protein structure in hand. We have very less option to follow.

The another thing i notice is that their are many methodology researcher developed and published in nice reputed journals but no single methodology became as popular as QSAR is.

Even though the people who are working with QSAR since last 20 years said that, QSAR not fulfill the expectation set by them, is reported in editorial letter (2007) of Journal of Chemical Information and Modeling.

In this concern, it would be my pleasure to know methodology which will became better substitute for QSAR.

To all, please give your opinion regarding R-group analysis(or Group based QSAR) and ligand based scaffold hopping.

Thanking you

Scaffold hopping is a very important method in ligand based design, not least because it might allow you to get outside some one else's patent space but retain the important binding properties of the molecule you start with. It also may be used for escaping ADME and tox liabilities.

The bottom line is that all the major pharma.companies are doing this and, whilist I don't have any figures for a success rate, I'm sure it is a more useful tool in pragmatic drug design than standard QSAR, which quite frequently tells you no more than you knew already and is best used (IMHO) to fill holes in your SAR and patent scope.

R-Group analysis is of course very important and most medicinal chemists themselves are skilled at doing this manually as a matter of course in their day to day research. So it is is rare (but by no means unheard of) that a computational R Group analysis can tell them something they didn't spot themselves in the structure/activity data

Pay attention to sophisthods, in general they do not work:

http://pubs.acs.org/doi/abs/10.1021/ci0500177

Excuse my error I intended to say 'sophisticated methods' 'sophistods' doe not mean anything....the attached link is in any case in my opinio extremely important on a general epistemological perspective, not only for medicinal chemistry.

I had a look again and I can provide you with two publications to the topic.

In the PPT slide deck there is also the citation to Warren et al.

http://shoichetlab.compbio.ucsf.edu/publications/jm060999m.pdf

http://www.mmsconferencing.com/pdf/htmc/c.edge.pdf

I wonder if there is some confusion in the original question with the word discovery. When I use the word discovery I mean given one or many active molecules the discovery of new active molecules. For this shape based methods (ROCS), electrostatic methods (Fieldscreen or Eon) and pharmacophore methods like Catalyst have been shown to be successful historically. By the way shape and pharmacophore methods can use multiple molecules. The difficulty is alignment and not over parametrizing the query. Electrostatic methods have the highest probability of identifying scaffold hops. If you are asking about ligand based lead optimization then the most widely used methods is QSAR. CoMFA, a variant of Catalyst the name I can't remember, and LIE have been used. Each method has well documented liabilities (mostly poor applicability outside the domain of knowledge). The biggest liability for ligand based lead optimization is the lack of bounding. With structure-based design you can hypothesize that adding a larger R-group that collides with the protein is a bad idea. There is no equivalent for ligand-based lead opt and I don't see a solution arising. Without exploring unknown space with synthesis and testing you cannot know when or where changes will be successful.

Thank you very much for your kind post.

@ Andreas, the ppt shared by you is really worthy. The Daylight based substructure technique sounds worthy.

Generally researcher follow docking studies after phrmacophoric screening, to give more strengthen support for future wet lab studies.

What will be the case if we have no protein structure?

which methodology we can adopt when we have no protein structure, for further validation? especially apart from ADME prediction.

thank you

Biological evaluations are the ultimate.

Yes, it is mandatory to go for biological evaluation.

But how to decrease number of molecules to be screened for final biological evaluation without docking?

You should consider the myriad of molecular similarity methods which have been succssfully applied to Ligand-based drug discovery and which do not require a protein structure.

I recommend LINGOsim because it is the fastest and COSMOsim-3D because it is the most precise.

In addition, also consider visual inspection of the top compounds to check if they are Med Chem friendly e.g remove compounds with reactive functional groups.

Thank you Michael, Thank you Grace.

Well, a refined computational protocol can take advantage of several theoretical approaches: (i) docking, for instance, with bindsurf (http://bio-hpc.eu/software/bindsurf/) (ii) molecular dynamics (gromacs...) and eventually hybrid QM/MM (NWChem, ORCA...). Notice that all proposed software are free of charge for academic use. Alternatively, Schrödinger has a nice package called "Small Molecule Drug Design" (https://www.schrodinger.com/smdd/), but at $$. Hope that helps.

Docking at the known inhibitor bound crystallographic structure may be helpful.

Thank you jose and apurba,

but actually i started this discussion to know technique we can adopt for computational drug discovery, when we dont have protein structure.

For docking we need to have a crystallographic structure.

If you want to decrease the number of molecules to test biologically, i would suggest the following, not necessarily in this order:

-filter your molecules to remove molecules with instable/reactive functional groups and/or molecules that are synthetically unfeasible.

-clustering (in any property space, with any clustering method): Often, virtual hits are very similar, (methyl vs ethyl rest) Its better (IMHO) to initially test a diverse set of molecules than very similar molecules. Clustering can help you to pick the most diverse subset.

-visual inspection by chemist (cherry picking): Often, cost becomes a factor, so discuss with the synthetic chemist which molecules are easiest to make/ easiest to purify.

-check literature / patents whether the structure is "new" at all. Often you will find that active molecules have been reported to be active on other targets, or were just patented by big pharma en mass.

hope this gives you some new ideas

Apart from the protocols like QSAR and 3D pharmacophore development, the ligand based study includes QM/MM calculations for the ligands with the protein, the binding free energy calculations, MMGBSA/PBSA studies of that particular ligands etc which can help to finalise the molecules for biological evaluation.

FEP calculations will indeed be beneficial for the selection of molecules.

Apart these, you can extend your work based on cores and protocores mutational studies which can be termed as scaffold hopping.

you should try fragments based approach like if you have active compound for the given targets try to figure out which fragments are always presents and what fragments are absent in active as well as inactive compound you can use MCS approach also

Thank you Julian priyanka and sameer.

@ julian, can you suggest any free software for clustering.

commercial (www.schrodinger.com) or free (Autodock or Gold) docking programs.

Also interesting, but still a beta version: www.drugable.com

Greetings

CL

Thank you Christoph and Duangrudee

Finding new molecules with interesting biology using ligand information only:

1D methods like property matching.

2D or graph-based methods (fingerprints, substructure searching).

3D methods ( shape matching, electrostatic potential matching).

You can do all this with tools from OpenEye, which are free to qualified academics.

Discovering new drugs on the other hand is a much more difficult problem...

but i want to build the ligand traing set. how is it possible?

i have two questions. how can i differentiate all the ligand into training set and test set?

and second question is after pharmacophore generation the cost difference is very less. what should be the probable problems?