From owner-chemistry &$at$& ccl.net Wed Jun 28 02:22:30 1995 Received: from dingo.cc.uq.oz.au for M.Dooley-0at0-mailbox.uq.oz.au by www.ccl.net (8.6.10/930601.1506) id CAA03229; Wed, 28 Jun 1995 02:18:46 -0400 Received: from localhost by dingo.cc.uq.oz.au with SMTP id AA01681 (5.67a/IDA-1.5 for ); Wed, 28 Jun 1995 16:18:44 +1000 Date: Wed, 28 Jun 1995 16:18:37 +1000 (GMT+1000) From: Michael Dooley X-Sender: ddmdoole[ AT ]dingo.cc.uq.oz.au To: Compuational Chemistry List Subject: Re:POSTED RESPONSES: Quantitative assessment of novel ligands Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Dear Netters, Thanks to everyone who responded to my query >I'm searching for a protocol or recipe of 'vital statistics' to assess >novel ligands that i've designed, so that i'll have firm arguments to >convince organic chemists to synthesise a molecule. I'm using biosym >software + intuition to de novo design ligands for a target with good >x-ray data and also have access to Oxford Molecular software. If it >might not be possible to roughly predict binding constants for novel >compounds that differ greatly from known ligands, are there methods to >predict the effects of modifications to known ligands. I'm aware of >scoring functions such as Bohm Comput-aided mol des (1994) 243, but I'm >not sure if this would be appropriate. Eagerly awaiting any suggestions. ************************************************************************* Michael, We have done some preliminary work on assessing binding energies as calculated from a series of molecular dynamics runs of an inhibitor bound to the active site of Influenza sialidase. This can be found in J Med Chem 37, 616 (1994) and the work was done by a colleague Neil Taylor. A more comprehensive paper is at the moment being reviewed for publication in J Comp Chem, but I'm not sure when that will published. I have adapted the protocol Neil developed to several other Sialidase's and have obtained reasonable results that enable me to at least say whether a compound is worth proceeding with. The protocol involves a series of molecular dynamics runs on the inhibitor docked into the active site of the protein/enzyme of interest and then determining the pairwise non-bonded interaction energy between the inhibitor and the protein. This enables me to determine a set relative binding energies. i.e. compound A > compound B >> compound c = compound d. Both Neil and I will hopefully be presenting results at the drug-design conference later in the year in Cairns. I forgot to mention we used Biosym's InsightII and Discover for the calculations and viewing of the results. Bye Jeff. -- Jeff Dyason Department of Medicinal Chemistry, Victorian College of Pharmacy, Ph: 61 3 9903-9110 Monash University, Parkville, Fax: 61 3 9903-9582 Melbourne Victoria Australia e-mail: jcd (+ at +) vcp.monash.edu.au ****************************************************************************** Dear Michael Dooley. As you are undoubtedly aware there are no simple solutions to your problem. And the problem is not unique to you, everyone in drug design faces it. I have just a few comments to make: We have implemented Boehm's LUDI scoring algorithm (in Sybyl), and find it somewhat unsatisfactory because it is overly sensitive to the positions of the atoms, and the hydrogens in particular. Also, one could wish to make a more sophisticated distinction between various types of hydrogen bonds. An inherent deficiency in considering the ligand/protein complex is that the ligand/water (and protein/water) complexes are equally important. To my knowledge the only proper way to treat this is by means of free energy perturbation calculations (simulations). Which means hard work or (and ?) good luck to get decent results. Chemical intuition is likely to be your best tool. (although an analysis tool like Goodford's GRID is helpful) But you and your synthetic chemists have to accept that our ability to predict is not perfect, - it is probably more like that of the meteorologist making his weather forecast: better than random. Sorry, if this sounds awfully pessimistic. Cheers, Leif Norskov Novo Nordisk A/S Copenhagen Denmark lnl.,at,.novo.dk ******************************************************************************** In response to your question - 'what molecular parameters can be generated to convince an organic chemist to synthesise a molecule you have designed?' Before embarking on this process you should also view the question from the organic chemists point of view. He will be considering your compound and prioritising it with other compounds he is already synthesising and will probably make this assessment based on i) synthesisability (a major problem with de novo designed ligands) ii) the state of the project (is it early days and leads are scarce, or is it towards the end of the project when fine tuning is the focus?) iii) how far is the structure away from the known SAR iv) in the light of ii and iii what are the chances of success and are the risks worth taking. There are probably other questions, but I doubt whether any other parameters will make a great difference to your ability to persuade a busy organic chemist to synthesise your compounds. Presumably you are convinced that the molecules you have designed are worth making, therefore your best approach is to present your argument based on how you have designed the molecules, but most important of all, you should have your chemist collaborators on board at an early stage in the modelling and involve them as much as possible in the process. Along with the basic tenet of modelling: 'No modelling without experimentation' (A. Vintner) we should possibly add: 'No synthesis without dialog' ?! If you are not convinced yourself that these molecules are worth making and that their synthesis will answer this question, then you need to get into activity prediction, i.e. QSAR using MOPAC generated parameters, principal component analysis etc, or using Comfa (Tripos). These are not trivial exercises but do give a prediction that may help to decide whether the molecules are worth synthesising. Good luck anyway! Chris Snell Molecular Modelling and Computational Chemistry Sandoz Institute for Medical Research 5 Gower Place London WC1E 6BN E-Mail snell "at@at" sandoz.com Tel 0171-333-2165 Fax 0171-387-4116 ******************************************************************************* Michael, A program from edusoft called HINT (Hydrophobic Interactions) will give a rough estimate for the binding constant of a ligand/protein complex. The www address is a follows: http://www.i2020.net/edusoft/hint.html The nice thing about this program is the "pretty" pictures that can be generated in order to "validate" your claim for ligand synthesis. In my experience most synthetic organic chemists seem to love color graphics and are usually willing to bend over backwards to assist you with your problem once you provide them with some justification. The software currently interfaces with InsightII, Sybyl, and Chem-X. Also, QSAR analysis can carry alot of weight with synthetic organic chemists (a standard module in most software packages). I would suggest that COMFA QSAR would carry more weight than standard QSAR. Furthermore, HINT can be used as an additional COMFA field with QSAR within Sybyl. If I can provide further assistance, please drop me a line. As a side note, do you know how difficult it is to obtain a post-doctoral position in your neck of the woods? I hope that I have been of some help, Shawn Feaster University of Iowa Dept. of Chemistry feaster \\at// tessa.iaf.uiowa.edu ****************************************************************************** There are probably over a hundred citations for the use of CoMFA as a tool for predicting biological activities (conversely I know of none for BioSym or Oxford Molecular save that of Boehm and Richards, the developers). If you are involved with known receptor structures, some of the work by Garland Marshall and his collaborators, for example in JACS, might be relevant. Sorry I don't have specific referebnces handy -- yet another biased developer, Dick Cramer cramer-0at0-xhost3.tripos.com ****************************************************************************** A variety of QSAR techniques have been developed for the purpose of predicting the activity of ligands. CoMFA, comparative molecular field analysis, is a relatively new technique which was developed by Dick Cramer of Tripos. There is a catch. Tripos holds a patent on the method and you need a license for Sybyl for the job. CoMFA involves the following steps: 1) Generate a set of molecules (learning set + set of new molecules) and superimpose them. The hard step is deciding on a reasonable conformation and determining which atoms to use in the superposition. You also need to calculate the charges on the molecule. MOPAC with the AM1 Hamiltonian works well. 2) You save the set of molecules with charges in a molecular database. 3) Each molecule in the database is imbedded in a three-dimensional grid. A charged probe molecule is placed at each point in the grid and the interaction energy between the probe and the ligand is calculated at each grid point. The three-dimensional matrix of energy values, one per grid point, is called a field. Tripos' software divides the energy calculation into two parts: a steric portion based on van der Waals interactions and an electrostatic part. 4) In the last step, you correlate the biological or chemical activity of the ligands with the field values. In effect, each energy at each grid point corresponds to an "independent variable" so the number of independent variables far exceeds the number of molecules in the database. Hence the method of partial least squares (PLS) is required to construct the model. In the PLS algorithm, linear combination(s) of the field values is(are) calculated which best explain the data. These linear combinations are the new independent variables and are called components. 5) With these components in hand, you can predict the activity of the new ligands. I don't guarantee success. The Martin group at Abbott and the Kubinyi group at BASF have had extensive experience with the method. CoMFA did quite well in modelling logP of amino acids. I am finishing up a CoMFA analysis of inhibitors of a serine protease for the Hansch group at Pomona. Wayne Steinmetz ******************************************************************************** I would be very interested to receive a summary of replies. We face similar problems for the assessment of novel ligands. We have found variants of Bohm's scoring function useful in conjunction with simple simulation protocols to check the ligand binds as expected. The problem is that if your set of designs bind differently and/or have very different chemistry, one needs to have alot of faith in the ability of current methods to differentiate between them. However reparametrising Bohm-like scoring functions to agree with experimental data for the receptor you're interested in should be effective. Of course this isn't always possible. There's a recent paper by people at Merck who correlated interaction energies with binding affinities when designing HIV protease inhibitors and they got nice results. with thanks chris Chris Murray | Proteus Molecular Design Ltd., | Tel: 01625-500555 Lyme Green Business Park, | Fax: 01625-500666 Macclesfield, Cheshire, | Email: C.W.Murray' at \`proteus.co.uk SK11 0JL, UK | ********************************************************************************* Dear Michael, According to your question in CCL about tools for the ligands design, I want to turn your attention to the program Apex-3D integrated with insightII (BIOSYM). It allows among other things to build 3D QSAR models and predict binding constants for novel compounds. A short description of the methods incorporated into Apex-3D can be found in the http://www.dcl.co.il/apex3d.html. More detailed description is in the Apex-3D Manual available from BIOSYM. Please feel free to contact me for additional information. Sincerely yours Boris Vesterman *---------------------------------------------------* | Boris Vesterman, Ph.D. | | | | DCL Systems International Ltd. | | 20 Galgalei Haplada St. Herzlia Industrial Area | | P.O.B. 544 Herzlia 46105 Israel | | Phone: 972-9-584684 | | Fax: 972-9-543917 | | E-mail: boris[ AT ]dcl.co.il | *---------------------------------------------------* ********************************************************************************* If you have a set of related molecules you might be successful with CoMFA. The best references are to Garland Marshall in J. Med. Chem. His group did several different test cases. You could do the calculations with Oxford Molecular ASP/TSAR. Let me know what other answers you get. (- at -) (- at -) (- at -) (- at -) (- at -) (- at -) (- at -) (- at -) (- at -) (- at -) (- at -) Yvonne Martin, Senior Project Leader $#at#$ Computer Assisted Molecular Design Project ()at()()at()()at()()at()()at()()at()()at()()at() ()at() D-47E, AP10 2fl (-(at)-) (-(at)-) Abbott Laboratories #*at*# #*at*# 100 Abbott Park Road &$at$& &$at$& &$at$& &$at$& &$at$& &$at$& &$at$& &$at$& &$at$& &$at$& Abbott Park, IL 60064-3500 Phone: 708 937-5362 FAX: 708 937-2625 yvonne.martin -x- at -x- abbott.com Andy McCammon wrote a nice review article: (1) Straatsma, T. P.; McCammon, J. A., "Theroetical Calculations of Relative Affinities of Binding," Methods in Enzymology 1991, 202, 497-511. yvonne.martin[ AT ]abbott.com ****************************************************************************** ***************************************************************************** I further e-mailed Boris (below) and his speedy reply is printed at bottom. Thanks Boris! >Dear Boris, > Thankyou for your fast reply to my query on quantit. asess. >novel ligands. On the subject of Apex-3D, a member of our group >has investigated the application of this software to our problem. >This avenue is still under investig. but this is our No 1 problem: > we have many crystal structures of target and ligand and so have very >good information on overlays of different ligands. Apex-3D wants to >manipulate our data set away from reality to suit its own ideas on how >the ligands should align, which seems like a backwards step! Any ideas? >Cheers >Michael Dooley Michael, I am familiar with this problem. When Apex-3D was created most of the attention was paid to the "black box" approach, when no X-ray data are available for enzyme-ligand complex. Unfortunately in case when you have additional information about crystal structures of target and ligand, Apex-3D still tries to generate its own superposition hypothesis. You can reduce this list of different biophores by selection of most appropriate Task Definition parameters and tolerances. I mean the following: to describe atom_class's and atom_property's for the specific ligand groups, reduce tolerance ( as narrow as possible ) to decrease the number of possible alignments. May be in some cases it is not so simple, and I'll be happy to answer any questions you have. BTW, this autumn Apex-3D new version will be shipped ( integrated with InsightII v.95.0. There is a set of tools for much flexible task definition and user control of biophore and model selection. A special tool that is currently under development can help to solve your problem. It allows to create a biophore from a user defined superposition. All your suggestion in this direction are highly appreciated. Best Regards Boris Vesterman *---------------------------------------------------* | Boris Vesterman, Ph.D. | | | | DCL Systems International Ltd. | | 20 Galgalei Haplada St. Herzlia Industrial Area | | P.O.B. 544 Herzlia 46105 Israel | | Phone: 972-9-584684 | | Fax: 972-9-543917 | | E-mail: boris <-at-> dcl.co.il | *---------------------------------------------------* ****************************************************************************