Re: CCL: Computational drug design blues

From owner-chemistry@ccl.net Fri Sep 16 02:08:24 2005 From: "CCL" To: CCL Subject: CCL: Computational drug design blues Message-Id: <-29206-050915234211-27160-0x46I4zu8jI1TAcCXAQwZg**server.ccl.net> X-Original-From: Dave Winkler Content-Type: multipart/alternative; boundary="============_-1085285826==_ma============" Date: Fri, 16 Sep 2005 12:11:06 +1000 Mime-Version: 1.0 Sent to CCL by: Dave Winkler [dave.winkler**csiro.au] Replace ** with ** to recover original email address. --============_-1085285826==_ma============ Content-Type: text/plain; charset="us-ascii" ; format="flowed" > >My argument is that the practicality issue has so many variables >that it is very difficult to implement in an algorithmic fashion. I >also am quite confident in saying that no existing program has made >any realistic attempt to model practicality at the level of the >bench chemist. So, without this kind of input, I do not believe >that any existing program can make a meaningful assessment of >synthetic feasibility (as distinct from strategic efficiency, which >has been addressed to a fair degree). > >Dr. Philip G. Hultin I don't think this is strictly true, depending on how you define a 'program'. I remember a paper by a Japanese worker recently that took a very empirical and practical approach. He collected together a large set of molecules and asked a large number of competent synthetic chemists to score the molecules in terms of 'drug likeness' and synthetic feasibility. He used the data to generate a neural net model that apparently had some degree of predictivity and usefulness. -- Cheers, Dave Prof. David A. Winkler FRACI CChem CPChem Email: dave.winkler**csiro.au CSIRO Centre for Complexity in Drug Design Voice: 61-3-9545-2477 Senior Principal Research Scientist Fax: 61-3-9545-2561 CSIRO Molecular and Health Technologies or 61-3-9545-2446 Private Bag 10,Clayton South MDC 3169 http://www.csiro.au Australia http://www.molsci.csiro.au Join us at the 11th Asian Chemical Congress in Seoul, August 24-26 2005 (http://www.11acc.org/), Pacifichem 2005 in Honolulu December 15-20 (http://www.pacifichem.org), MM2006 in Perth, Australia 19-22 April 2006 (http://www.nanochemistry.curtin.edu.au/conferences/mm_2006.cfm), Drug Design Amongst the Vines, Hunter Valley, Australia 3-7 December 2006 (http://www.ddconference.com) --============_-1085285826==_ma============ Content-Type: text/html; charset="us-ascii" Re: CCL: Computational drug design blues

My argument is that the practicality issue has so many variables that it is very difficult to implement in an algorithmic fashion. I also am quite confident in saying that no existing program has made any realistic attempt to model practicality at the level of the bench chemist. So, without this kind of input, I do not believe that any existing program can make a meaningful assessment of synthetic feasibility (as distinct from strategic efficiency, which has been addressed to a fair degree).

Dr. Philip G. Hultin

I don't think this is strictly true, depending on how you define a 'program'. I remember a paper by a Japanese worker recently that took a very empirical and practical approach. He collected together a large set of molecules and asked a large number of competent synthetic chemists to score the molecules in terms of 'drug likeness' and synthetic feasibility. He used the data to generate a neural net model that apparently had some degree of predictivity and usefulness.

--

Cheers,
Dave

Prof. David A. Winkler FRACI CChem CPChem    Email: dave.winkler**csiro.au
CSIRO Centre for Complexity in Drug Design        Voice: 61-3-9545-2477
Senior Principal Research Scientist                Fax: 61-3-9545-2561
CSIRO Molecular and Health Technologies     or 61-3-9545-2446
Private Bag 10,Clayton South MDC 3169 http://www.csiro.au
Australia

http://www.molsci.csiro.au

Join us at the 11th Asian Chemical Congress in Seoul, August 24-26 2005 (http://www.11acc.org/), Pacifichem 2005 in Honolulu December 15-20 (http://www.pacifichem.org), MM2006 in Perth, Australia 19-22 April 2006 (http://www.nanochemistry.curtin.edu.au/conferences/mm_2006.cfm), Drug Design Amongst the Vines, Hunter Valley, Australia 3-7 December 2006 (http://www.ddconference.com)

--============_-1085285826==_ma============-- From owner-chemistry@ccl.net Fri Sep 16 03:41:19 2005 From: "CCL" To: CCL Subject: CCL: W:gaussian Message-Id: <-29207-050916033844-26255-rFRPexv7AytkJeV34UV2SA : server.ccl.net> X-Original-From: Olaf Othersen Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=ISO-8859-1; format=flowed Date: Fri, 16 Sep 2005 08:46:58 +0200 MIME-Version: 1.0 Sent to CCL by: Olaf Othersen [olaf.othersen : chemie.uni-erlangen.de] Replace strange characters with an : sign to recover email address. CCL wrote: >Sent to CCL by: "Rafael Carlos Garcias" [rafel.garcias|-|uib.es] >Dear CCL members, >I have a problem with my force calcultion. Once I have optimized an structure at B3LYP 6-31+g* level I start the force calculation by the keyword freq in the optimized structured obtained, but the problem is that the calculation finish because I have spent all the calculation time avaliable. I've tried to restart the freq calculation by the *.chk but it doesn't work. How can I restart the freq calculation or how can I reduce the computational time without changing the basis set? >Thank you> > > > > Hello Rafael, if you do not need the data for raman spectroscopy use freq=noraman. This will save you a lot of time in the frequency calculation. Olaf From owner-chemistry@ccl.net Fri Sep 16 05:06:15 2005 From: "CCL" To: CCL Subject: CCL: transition metals Message-Id: <-29208-050916044431-13636-RRMErCD2jcYnjZ0vpqYuuA]~[server.ccl.net> X-Original-From: Marcel Swart Content-Type: multipart/alternative; boundary=Apple-Mail-5-376626299 Date: Fri, 16 Sep 2005 10:43:17 +0200 Mime-Version: 1.0 (Apple Message framework v622) Sent to CCL by: Marcel Swart [m.swart]~[few.vu.nl] Replace strange characters with an ]~[ sign to recover email address. --Apple-Mail-5-376626299 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset=WINDOWS-1252; format=flowed Dear all, I would suggest the use of OPBE for transition metal complexes. Not only is it a pure GGA functional, e.g. computationally efficient so you can treat larger systems efficiently, it also has the additional advantage that it does not suffer > from spin contamination as much as hybrid functionals do. Apart from that, it gives very good performance, not only for the spin state energies of iron complexes, but also for other properties. See: M. Swart, A.W. Ehlers and K. Lammertsma "Performance of the OPBE exchange-correlation functional" Mol. Phys., 2004, 102, p. 2467-2474 M. Swart, A.R. Groenhof, A.W. Ehlers and K. Lammertsma "Validation of exchange-correlation functionals for spin states of=20 iron-complexes" J. Phys. Chem. A, 2004, 108, p. 5479-5483 The OPBE functional is available in the ADF program, and should be available in other computer codes as well, by using the combination of OPTX exchange with PBE correlation. Marcel. On Sep 15, 2005, at 3:31 PM, CCL wrote: > > Sent to CCL by: Victor Nemykin [victor_nemykin---yahoo.com] > Hello Goedele: > You raised an interesting question, which actually has no > straightforward answer. Which property (properties) you would like to > calculate? I will choose B3P86 or B3LYP coupled with 6-31G(transition > metal)/6-31G(d) (other atoms) or Wachters (transition metal, > represented as 6-311+G in Gaussian)/ 6-31G(d) (other atoms) for > geometry optimization because hybrid EC functionals usually gives a > slightly better geometry as compared to the pure ones. Again, I am a > Gaussian user and people who are using ADF can argue this point in > favor of pure DFT EC functionals, for instance BP86. If you would like > to calculate vertical excitation energies, it seems that both hybrid > (B3P86, B3LYP) and pure (BP86) EC functiolans gives a reasonable > results in many cases (test also PBE1PBE). For Mossbauer parameters > calculations, I would recommend o use B3P86 or B3LYP, while EPR > parameter calculations usually go better with pure EC functionals. Of > course the choice of BS/EC functional is also depends on the size of > the molecule you would like to calculate because in general pure EC > functionals are faster as compared to hybrid ones. You can get some > very general ideas from my web-site, but probably if you will provide > me more information about the calculations you are interesting in, I > can give you a better advise=85 > > Sincerely, > Victor =96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96= =96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96 dr. Marcel Swart Theoretische Chemie Vrije Universiteit Amsterdam Faculteit der Exacte Wetenschappen De Boelelaan 1083 1081 HV Amsterdam The Netherlands Tel +31-(0)20-5987619 Fax +31-(0)20-5987629 E-mail m.swart]~[few.vu.nl Web http://www.few.vu.nl/~swart =96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96= =96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96 --Apple-Mail-5-376626299 Content-Transfer-Encoding: quoted-printable Content-Type: text/enriched; charset=WINDOWS-1252 Dear all, I would suggest the use of OPBE for transition metal complexes. Not only is it a pure GGA functional, e.g. computationally efficient so you can treat larger systems efficiently, it also has the additional advantage that it does not suffer > from spin contamination as much as hybrid functionals do. Apart from that, it gives very good performance, not only for the spin state energies of iron complexes, but also for other properties. See: M. Swart, A.W. Ehlers and K. Lammertsma "Performance of the OPBE exchange-correlation functional" Mol. Phys., 2004, 102, p. 2467-2474 M. Swart, A.R. Groenhof, A.W. Ehlers and K. Lammertsma "Validation of exchange-correlation functionals for spin states of iron-complexes" J. Phys. Chem. A, 2004, 108, p. 5479-5483 The OPBE functional is available in the ADF program, and should be available in other computer codes as well, by using the combination of OPTX exchange with PBE correlation. Marcel. On Sep 15, 2005, at 3:31 PM, CCL wrote: Sent to CCL by: Victor Nemykin [victor_nemykin---yahoo.com] Hello Goedele: You raised an interesting question, which actually has no straightforward answer. Which property (properties) you would like to calculate? I will choose B3P86 or B3LYP coupled with 6-31G(transition metal)/6-31G(d) (other atoms) or Wachters (transition metal, represented as 6-311+G in Gaussian)/ 6-31G(d) (other atoms) for geometry optimization because hybrid EC functionals usually gives a slightly better geometry as compared to the pure ones. Again, I am a Gaussian user and people who are using ADF can argue this point in favor of pure DFT EC functionals, for instance BP86. If you would like to calculate vertical excitation energies, it seems that both hybrid (B3P86, B3LYP) and pure (BP86) EC functiolans gives a reasonable results in many cases (test also PBE1PBE). For Mossbauer parameters calculations, I would recommend o use B3P86 or B3LYP, while EPR parameter calculations usually go better with pure EC functionals. Of course the choice of BS/EC functional is also depends on the size of the molecule you would like to calculate because in general pure EC functionals are faster as compared to hybrid ones. You can get some very general ideas from my web-site, but probably if you will provide me more information about the calculations you are interesting in, I can give you a better advise=85 Sincerely, Victor = Helvetica=96=96=96=96=96= =96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96= =96=96=96=96=96=96=96=96=96=96=96=96=96=96 = Papyrusdr. Marcel Swart = Papyrus = OsakaTheoretische Chemie Vrije Universiteit Amsterdam Faculteit der Exacte Wetenschappen De Boelelaan 1083 1081 HV Amsterdam The Netherlands Tel +31-(0)20-5987619 Fax +31-(0)20-5987629 E-mail m.swart]~[few.vu.nl Web http://www.few.vu.nl/~swart = Helvetica=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96= =96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96=96 --Apple-Mail-5-376626299-- From owner-chemistry@ccl.net Fri Sep 16 05:58:21 2005 From: "CCL" To: CCL Subject: CCL: W:Disclose your data, or not publish ! Message-Id: <-29209-050916055634-3156-i39Kd5k4sDCNONcpxKeGKg%server.ccl.net> X-Original-From: "Dr. Csaba Hetenyi" Content-Type: TEXT/PLAIN; charset=US-ASCII Date: Fri, 16 Sep 2005 11:09:27 +0200 (CEST) MIME-Version: 1.0 Sent to CCL by: "Dr. Csaba Hetenyi" [csaba%ovrisc.mdche.u-szeged.hu] Replace strange characters with an % sign to recover email address. Dear Dr. Hawkins, > docking tools used in the study are unsupportable. Consequently > providing the compound sets used in such studies should be mandatory. Naturally, I have never written anything against inclusion of full data in a publication. However, RMSD calculation and validation with the crystallographic ligand position is obligatory too. Docking methods can be structurally validated with RMSD-s, this is what i mean. Docking methods are generally validated in this way. Exceptions may be found in the literature: it depends on the editors and referees: what they accept for publication... > As to judging the quality of a scoring function by correlation with > binding affinity, no scoring function available in the popular docking > tools in current use shows any meaningful correlation with binding > affinity across diverse compound classes. Are you suggesting that any > manuscript using these tools in any context should be denied > publication? A consistent correlation between binding energy and score Sorry to say, that you have misinterpreted my words in a negative way. As a matter of fact, there was nothing to misinterprete in those few lines. I simply stated, that comparison with experimental delta G values (binding affinities in other words) is a possibility for validation. Concerning the quality of scoring methods, for a comprehensive table on them, please check e.g. Krammer et al. J. Mol. Graph. Model. (2005) 23 395-407. There are numerous other papers available on the topic of scoring functions... It should be emphasized, that this field is under heavy research, fortunately a lot of studies coming out monthly with new approaches and better correlations. Best wishes, Csaba Hetenyi From owner-chemistry@ccl.net Fri Sep 16 09:27:47 2005 From: "CCL" To: CCL Subject: CCL: Anharmonic frequences in Gaussian Message-Id: <-29210-050916055657-3202-MEAxJg+sxvq2MVgktdvekA : server.ccl.net> X-Original-From: Alejandro Pedro Ayala Content-Disposition: inline Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset=ISO-8859-1 Date: Fri, 16 Sep 2005 01:57:09 -0700 Mime-Version: 1.0 Sent to CCL by: Alejandro Pedro Ayala [ale.p.ayala : googlemail.com] Replace strange characters with an : sign to recover email address. Hi, I am trying to perform some frequency calculations in Gasussian using the anharmonic keyword. Since they are highly time consuming, I would like to estimate how many time these calculations need. I looking to the output file I saw the cycles through the different modules Link 106 -> Link 301 ..... Link 716, etc. So, my question is: Is it possible to know a priory how many of these cycles the calculation will need? Regards, Alejandro Ayala From owner-chemistry@ccl.net Fri Sep 16 11:25:51 2005 From: "CCL" To: CCL Subject: CCL: W:Disclose your data, or not publish ! Message-Id: <-29211-050916110541-31618-ntk3ZLaJCaLFDVX5zajMBA-#-server.ccl.net> X-Original-From: Paul Hawkins Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII; format=flowed Date: Fri, 16 Sep 2005 08:15:45 -0600 MIME-Version: 1.0 Sent to CCL by: Paul Hawkins [phawkins-#-eyesopen.com] --Replace strange characters with the "at" sign to recover email address--. CCL wrote: >Sent to CCL by: "Dr. Csaba Hetenyi" [csaba%ovrisc.mdche.u-szeged.hu] > >Replace strange characters with an -#- sign to recover email address. >Dear Dr. Hawkins, > > > >>docking tools used in the study are unsupportable. Consequently >>providing the compound sets used in such studies should be mandatory. >> >> > >Naturally, I have never written anything against inclusion of full data in >a publication. However, RMSD calculation and validation with the >crystallographic ligand position is obligatory too. Docking methods can be >structurally validated with RMSD-s, this is what i mean. Docking >methods are generally validated in this way. Exceptions may be found in >the literature: it depends on the editors and referees: what they >accept for publication... > > You are absolutely right. > > >>As to judging the quality of a scoring function by correlation with >>binding affinity, no scoring function available in the popular docking >>tools in current use shows any meaningful correlation with binding >>affinity across diverse compound classes. Are you suggesting that any >>manuscript using these tools in any context should be denied >>publication? A consistent correlation between binding energy and score >> >> > >Sorry to say, that you have misinterpreted my words in a negative way. >As a matter of fact, there was nothing to misinterprete in those few >lines. I simply stated, that comparison with experimental delta G values >(binding affinities in other words) is a possibility for validation. > > Quote Scoring can be validated with experimental delta G values too. If this validation is not done, of course, the manuscript should not be accepted. I take this to mean that if a scoring function does not show good correlation with exptl. binding affinities then work done with that scoring function should not be published since the scoring function is not validated. Did I misinterpret? >Concerning the quality of scoring methods, for a comprehensive table on >them, please check e.g. >Krammer et al. J. Mol. Graph. Model. (2005) 23 395-407. >There are numerous other papers available on the topic of scoring >functions... >It should be emphasized, that this field is under heavy research, >fortunately a lot of studies coming out monthly with new approaches >and better correlations. >Best wishes, >Csaba Hetenyi > > If we are going to play "Who reads more literature" I would suggest you read Chapter 3 of Virtual Screening in Drug Discovery, Alvarez and Shoichet, where the Vertex team present a very insightful analysis of the way in which dataset selection affects the apparent quality of scoring functions. Their conclusion is that none of the scoring functions examined show any robust correlation with binding affinity Since dataset selection showed a profound effect on the apparent quality of the scoring function the paper further underscores the necessity of making such data publicly available. The point that I am trying to make is that without access to the datasets used in VS evaluations of (docking) tools the publications cannot be reproduced. As such the publications are of limited value, as they do not allow for independent replication. Independent replication is an important part of doing science. Paul. From owner-chemistry@ccl.net Fri Sep 16 11:44:59 2005 From: "CCL" To: CCL Subject: CCL: Computational drug design blues Message-Id: <-29212-050916111230-32180-2fRcKLUMlH9tgF+JVYEASw\a/server.ccl.net> X-Original-From: "Phil Hultin" Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset="US-ASCII" Date: Fri, 16 Sep 2005 10:12:18 -0500 MIME-Version: 1.0 Sent to CCL by: "Phil Hultin" [hultin\a/cc.umanitoba.ca] --Replace strange characters with the "at" sign to recover email address--. Eugen Leitl's observations strike me as rather missing the point of the discussion. Behind the suggestion that databases exist for physical properties, toxicity etc is the assumption that every compound in the synthetic pathway is a known substance. This is demonstrably false by simple inspection of the literature - the majority of intermediates made during the synthesis of a new compound are themselves also new compounds. The question of assessing synthetic feasibility arises from de novo design of candidate structures, and only after they have been found to be novel compounds. Obviously, if the computational chemist suggests a commercially-available compound or one for which a synthesis has already been published, the "accessibility" of that compound is known and established. Even in this case, however, the properties of many of these substances are only incompletely catalogued in the literature. Finally, as a synthetic chemist who dabbles in computational chemistry, I can assure everyone that literature precedent is not a guarantee of identical chemo-, regio- or stereoselectivity when the published reaction is applied to a new molecule. Yes, IF one is simply repeating the exact reaction reported in the paper, one should get the same result, but again, we are talking about de novo prediction on new chemical substances here. You may or may not get the same levels of selectivity when the reaction is applied to your compound. The synthetic chemist can make educated guesses as to the breadth of applicability of any published method. The computer can also be programmed to do this, possibly using "expert system" methods etc. The question is how much effort is needed to make the prediction reliable? Isn't it simpler and probably cheaper to just use the expertise of the med chem. department in your company? Dr. Philip G. Hultin Associate Professor of Chemistry, University of Manitoba Winnipeg, MB R3T 2N2 hultin\a/cc.umanitoba.ca http://umanitoba.ca/chemistry/people/hultin -----Original Message----- > From: owner-chemistry\a/ccl.net [mailto:owner-chemistry\a/ccl.net] Sent: September 15, 2005 8:26 AM To: Hultin, Philip G. Subject: CCL: Computational drug design blues Sent to CCL by: Eugen Leitl [eugen a leitl.org] On Wed, Sep 14, 2005 at 05:23:25PM -0500, CCL wrote: > I note that the question of assessing the "ease of synthesis" has excited a > fair bit of comment. As I already pointed out, and as also stated by Gary > Breton, the organic chemistry community does not think it is actually > possible to meaningfully assess "ease of synthesis" by an algorithmic > approach. Algorithmic approach can mean many things. It could mean have the machine look up the structure in the supplier's catalogue, or a reaction database, which could result in a commercial product, or a literature citation. If the lead candidate is commercially available, or has been characterized in literature the problem is already (almost) solved. If this does not work, a substructure search would be in order. If this search provides no clear candidates, the system could try to reach close (in terms of synthetic steps) points in structure space, starting on the terra firma of the empirical island. If this still fails, it can propose a synthetic route from scratch (this is the last resort, and it will be likely not very good, or worse). We also have to look at the amount of chemical expertise of the user, from the worst case of naive user, to a reasonably competent chemist -- just not a synthetic chemist. Here many nonfertile branches can be pruned, before giving this into the hands of the (hopefully reasonably competent, and not just a bench monkey) synthetic chemist. The reasons for this might be a lack of good personnel, attempt to reduce costs, or just optimize the amount of compounds screened within a given cost envelope. Would it be effective, in the end? No idea. > Other issues that bear on the > practicality question are (not a comprehensive list): > > cost of raw materials > > cost of solvents > > cost of reagents, catalysts etc. This should be all on file, and preferrably directly accessible to the algorithm sorting the recommendations by the "practicality" rank. > > cost of disposal of waste materials Good point. Probably hard to obtain relevant data here. Ruthless cost-optimizers would just want to outsource the problem, preferrably to a place with less restrictive environmental regulations. > > health and lab safety hazards associated with reagents, solvents and > intermediate compounds Toxicity databases. There are even some de novo toxicity prediction tools, though probably not entirely reliable. > > level of selectivity actually attainable (stereoselectivity, > regioselectivity, chemoselectivity) If it's a literature reaction, the data is available. > > need for purification, particularly chromatography > > solubilities, boiling points and other physical parameters Physical properties database. Literature, too. > > expected thermodynamic and kinetic stability of intermediates > > scalability - can the proposed route be used only for milligram amounts or > can it be implemented on gram or kilo scales as needed? I do not think such practicality concerns arise already when one is just looking to obtain mg quantities for a bioassay, by the fastest and cheapest route available. It's just a candidate, after all, not even a lead. > The list could go on. The bottom line is that in order to determine whether > a proposed drug candidate is actually practical, you need to consider > numerous variables that cannot be reduced to a simple numerical scale. This > is where the experience and expertise of the bench chemist is essential. The bench chemist is ultimatively the person who's in charge of producing the compound, so the buck will eventually stop there. -- Eugen* Leitl leitl ______________________________________________________________ ICBM: 48.07100, 11.36820 http://www.leitl.org 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE From owner-chemistry@ccl.net Fri Sep 16 12:13:19 2005 From: "CCL" To: CCL Subject: CCL: Summary) Basis set for B3LYP Message-Id: <-29213-050916114739-19812-/4wk441S2sK4XEMIQteeNg]=[server.ccl.net> X-Original-From: "Telkuni Tsuru" Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset="WINDOWS-1252" Date: Sat, 17 Sep 2005 00:39:23 +0900 MIME-Version: 1.0 Sent to CCL by: "Telkuni Tsuru" [telkuni]=[venus.dti.ne.jp] --Replace strange characters with the "at" sign to recover email address--. Hello, CCLers. A few days ago, I sent the following question. ----- Original Question ----- Sent: Wednesday, September 14, 2005 2:30 PM Subject: CCL: Basis set for B3LYP > I like to know suitable basis set level for B3LYP on G98W. > > For example, MP2 calculation should not be carried out with > low level basis set(e.g. STO-3G). So I like to know lowest basis > set for B3LYP. Is "B3LYP/STO-3G" possible ? > > All responses, I will appreciate. > I will summarize them and send to CCL. I've received useful replies. In general, it is able to apply STO-3G to B3LYP. But the accuracy is another matter. When I calc a large system, I should use semiemperical PM3 or other method. Thank you very much for your cooperation. ///////////////// Summary ////////////////// ---1) from Dr. Marcel Swart --- Of course that's possible. Whether you get reliable results, is another thing.. In the end, you would like to have results that have converged with the basis set size, for which you will need 6-311++G(3df,2pd) or similar. Whether B3LYP still gives reliable results with this basis, depends critically on your system.. Of course, you could also do what most people do, and use the 6-31G* basis, and let B3LYP benefit from the cancellation of errors. ---2) from Drs. Goedele Roos --- Hi, a question, STO-3G is a minimal basisset, so I suppose you use it because your system is large. But why can't you use than MP3 or another semiemperical method? Because, in my opinion, you have to use at least a 3-21G basis set to have reliable results. ---3) from Andrew --- Dear Telkuni Tsuru, Just to let you know, B3LYP parameters were fitted using 6-31G* basis set. ---4) from Ph.D David Shobe --- B3LYP/STO-3G is possible, and G98W will calculate it for you without an error message. It's just not recommended because it gives you only very approximate energy differences (like HF/STO-3G). B3LYP/6-31G* (double zeta with polarization) is considered to be a "respectable" level of theory. ---5) from Dr. Lewars --- Hello, The B3LYP functional is normally used with the 6-31G* basis set or higher (e.g. 6-31G**, 6-311G*--of course there are other basis sets besides these Pople sets; the Dunning correlation-consistent sets are also popular). It is generally thought that the use of a smaller basis with a correlated method (DFT or correlated ab initio) is pointless, but I know of no definitive study of this. A rare case of B3LYP/3-21G is the calc. of IEs of carbenes by optimization at this level followed by B3LYP/6-31+G* single point energies: H M Muchall et al., Can J Chem, 1998, 76, 221. The 3-21G basis was better than the 6-31G* with CASSCF (not a DFT method) in a study of the Cope rearrangement: D A Hrovat et al., J Am Chem Soc, 1999, 121, 169. ---6) from Antonio Hernandez --- Dear Tsuru: In theory, you can use any contracted or uncontracted Gaussian basis set with B3LYP functional calculations. Nevertheless, mostly all these basis sets were optimized at the HF level of theory (or HF+correlation level). For consistency, I would prefer to use basis sets optimized at the Density Functional Theory level of theory, as those described by Salahub in: N. Godbout, D. R. Salahub, J. Andzelm and E. Wimmer, Can. J. Chem. 70 (1992) 560. You can get these at the website http://www.emsl.pnl.gov/forms/basisform.html under DZVP(DFT Orbital) , DZVP2(DFT orbital) and TZVP(DFT Orbital) or DGauss Polarized DFT Orbitals Basis Sets. I have personally obtained excellent results with this DFT + Salahub basis sets combination (see for instance: Theor. Chem. Acc. 106 (2001) 218). ---7) from Igor Avilov --- Dear Telkuni Tsuru, A couple of times I accidentally ran geometry optimizations of poly-atomic molecules with STO-3G basis (with B3LYP functional). The results were always meaningless. But already 6-31G did a good job, even for calculations of excited states. The choice of the basis set depends on molecular property in question and accuracy needed. Best regards, ---8) from JENS SPANGET-LARSEN --- Dear Tsuru, with respect to basis sets developed at the density functional level of theory, I draw your attention to the work of Frank Jensen: F. Jensen "Polarization Consistent Basis Sets. Principles." J. Chem. Phys. 115 (2001) 9113-9125 F. Jensen "Polarization Consistent Basis Sets. II. Estimating the Kohn-Sham Basis Set Limit." J. Chem. Phys. 116 (2002) 7372-7279 F. Jensen "Polarization Consistent Basis Sets III. The Importance of Diffuse Functions." J. Chem. Phys. 117 (2002) 9234-9240 ///////////////// Sincerely yours, ---------------------------------------------------- Telkuni Tsuru telkuni]=[venus.dti.ne.jp From owner-chemistry@ccl.net Fri Sep 16 14:07:38 2005 From: "CCL" To: CCL Subject: CCL: W:van der Waals radii Message-Id: <-29214-050916133759-27612-9NwgrM30ql1TaCwCoLOVlQ ~~ server.ccl.net> X-Original-From: "Gisele deFreitas Gauze" Sent to CCL by: "Gisele deFreitas Gauze" [giselegauze ~~ yahoo.com.br] --Replace strange characters with the "at" sign to recover email address--. I would like to know if is possible to determine (to calcule)the van der Waals radii using the Gaussian03 program. Regards for attention Gisele de Freitas Gauze giselegauze ~~ yahoo.com.br Maring-Paran-Brazil From owner-chemistry@ccl.net Fri Sep 16 14:49:17 2005 From: "CCL" To: CCL Subject: CCL: Computational drug design blues Message-Id: <-29215-050916144537-15771-j7cOgrO9jvniDVVet6Fa8A*_*server.ccl.net> X-Original-From: Eugen Leitl Content-Disposition: inline Content-Type: text/plain; charset=us-ascii Date: Fri, 16 Sep 2005 20:45:29 +0200 Mime-Version: 1.0 Sent to CCL by: Eugen Leitl [eugen*_*leitl.org] --Replace strange characters with the "at" sign to recover email address--. On Fri, Sep 16, 2005 at 10:12:18AM -0500, CCL wrote: > Sent to CCL by: "Phil Hultin" [hultin\a/cc.umanitoba.ca] > > Eugen Leitl's observations strike me as rather missing the point of the > discussion. Behind the suggestion that databases exist for physical > properties, toxicity etc is the assumption that every compound in the > synthetic pathway is a known substance. This is demonstrably false by Querying empirical databases was only the first step in the suggested algorithm. The original question was about predicting synthetic accessibility. (Btw, if anybody has access to Mini-Reviews in Medicinal Chemistry, Volume 4, No. 6, 2004; Predicting Synthetic Accessibility: Application in Drug Discovery and Development Pp. 681-692 J.C. Baber and M. Feher I would be very thankful for an electronic copy). Many properties of the target can be predicted (with varying degrees of accuracy) from its structure alone. Lipinski (drug-like and lead-like) rules do already do that. Toxicity prediction both from databases, similiarity to empirical structures and from scratch is not a new field. http://www.google.com/search?q=toxicity+prediction&start=0&start=0&ie=utf-8&oe=utf-8&client=firefox-a&rls=org.mozilla:en-US:official http://www.google.com/search?hl=en&lr=&client=firefox-a&rls=org.mozilla%3Aen-US%3Aofficial&q=property+prediction+drug&btnG=Search > simple inspection of the literature - the majority of intermediates made > during the synthesis of a new compound are themselves also new compounds. I agree that direct hits will be rare. Similiarities won't be so rare. De novo predictions will almost always predict a property, and most of them will be even not completely out of it. > The question of assessing synthetic feasibility arises from de novo design > of candidate structures, and only after they have been found to be novel > compounds. Obviously, if the computational chemist suggests a It would be too resource-intensive to check a structure for novelty at generation time. Furthermore it would be pointless: nobody has screened all commercially available and ever synthesized compounds against a specific target, and made the results freely available. > commercially-available compound or one for which a synthesis has already > been published, the "accessibility" of that compound is known and > established. When stepping through structure and conformation space, enumerating or randomly sampling cavity-constrained molecules the system does not know whether the matches it generates have been ever made by somebody, somewhere. It would not make sense to prune at that stage. Odds are overwhelming, anyway, that structure has never been made -- but that a similiar structure almost certainly has been, and that the synthetic route is in the literature, and that in some rare cases it even might be available off-the shelf in sufficient quantities for a screen. The structure generator will easily drown you in structures it thinks are neat, and ought to be screened. But in practice the guesses are not that good, and it is not cost-effective to synthesize even a fraction of the lead candidates. It would probably make more sense use that lead candidate as a starting point for a combinatorial library, and hope that the diversity ball it covers will contain real leads. As established by a honest, physical screen. > Even in this case, however, the properties of many of these substances are > only incompletely catalogued in the literature. The only relevant property in the drug discovery: does it bind, or not, is virtually guaranteed to not be on file. > Finally, as a synthetic chemist who dabbles in computational chemistry, I > can assure everyone that literature precedent is not a guarantee of > identical chemo-, regio- or stereoselectivity when the published reaction is > applied to a new molecule. Yes, IF one is simply repeating the exact > reaction reported in the paper, one should get the same result, but again, > we are talking about de novo prediction on new chemical substances here. > You may or may not get the same levels of selectivity when the reaction is > applied to your compound. All good points. > The synthetic chemist can make educated guesses as to the breadth of > applicability of any published method. The computer can also be programmed > to do this, possibly using "expert system" methods etc. The question is how > much effort is needed to make the prediction reliable? Isn't it simpler and > probably cheaper to just use the expertise of the med chem. department in > your company? Pharma industry is in a bind. The current rate of drug discovery does not return on investment, and yet the trend is still getting worse. The current situation is not sustainable, so both cost reductions and more accurate predictions methods will be required. Highly trained specialists and animal facilities are an expensive and/or scarce resource. There is no alternative to less people, and more models, both in machina, in vitro and in vivo, and screening automation. The good news is that the models never get worse. They only get better. -- Eugen* Leitl leitl ______________________________________________________________ ICBM: 48.07100, 11.36820 http://www.leitl.org 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE From owner-chemistry@ccl.net Fri Sep 16 18:47:12 2005 From: "CCL" To: CCL Subject: CCL: Computational drug design blues Message-Id: <-29216-050916170929-7654-WSzWhcrlkseIgu0M8TB9VQ:+:server.ccl.net> X-Original-From: Alejandro Pisanty Content-Type: TEXT/PLAIN; charset=US-ASCII Date: Fri, 16 Sep 2005 15:13:38 -0500 (CDT) MIME-Version: 1.0 Sent to CCL by: Alejandro Pisanty [apisan:+:servidor.unam.mx] --Replace strange characters with the "at" sign to recover email address--. Hi, congrats to CCLers for this discussion. Though a theoretical chemist by education and activity, my first lapidary reply to the original question was going to be, go back to your Morrison, "Organic Chemistry", and books with titles like "The Art of Organic Chemical Synthesis." Structures we can soooo easily calculate and model may turn out to have such impossible-to-achieve structural constraints, stereochemistry, etc., that make them essentially unachievable - or at amazingly low yields. Some of us started into chemistry knowing this was the greatest challenge... Alejandro Pisanty - a founder of the CCL, and admirative that it still is one of the healthiest online communities, after some 15 years of pioneering. Kudos to Jan Labanowski and all of us! . . . . . . . . . . . . . . . . . . . . . . . . . . Dr. Alejandro Pisanty Director General de Servicios de Computo Academico UNAM, Universidad Nacional Autonoma de Mexico Av. Universidad 3000, 04510 Mexico DF Mexico Tel. (+52-55) 5622-8541, 5622-8542 Fax 5622-8540 http://www.dgsca.unam.mx * ---->> Unete a ISOC Mexico, www.isoc.org Participa en ICANN, www.icann.org . . . . . . . . . . . . . . . . . . . . . . . . . . On Fri, 16 Sep 2005, CCL wrote: > Date: Fri, 16 Sep 2005 20:45:29 +0200 > From: CCL > Reply-To: CCL Subscribers > To: "Pisanty, Alejandro " > Subject: CCL: Computational drug design blues > > > Sent to CCL by: Eugen Leitl [eugen*_*leitl.org] > > --Replace strange characters with the "at" sign to recover email address--. > > On Fri, Sep 16, 2005 at 10:12:18AM -0500, CCL wrote: > > Sent to CCL by: "Phil Hultin" [hultin\a/cc.umanitoba.ca] > > > > Eugen Leitl's observations strike me as rather missing the point of the > > discussion. Behind the suggestion that databases exist for physical > > properties, toxicity etc is the assumption that every compound in the > > synthetic pathway is a known substance. This is demonstrably false by > > Querying empirical databases was only the first step in the suggested > algorithm. The original question was about predicting synthetic accessibility. > (Btw, if anybody has access to Mini-Reviews in Medicinal Chemistry, Volume 4, > No. 6, 2004; Predicting Synthetic Accessibility: Application in Drug Discovery > and Development Pp. 681-692 J.C. Baber and M. Feher I would be very thankful > for an electronic copy). > > Many properties of the target can be predicted (with varying degrees > of accuracy) from its structure alone. Lipinski (drug-like and lead-like) rules > do already do that. Toxicity prediction both from databases, similiarity > to empirical structures and from scratch is not a new field. > > http://www.google.com/search?q=toxicity+prediction&start=0&start=0&ie=utf-8&oe=utf-8&client=firefox-a&rls=org.mozilla:en-US:official > http://www.google.com/search?hl=en&lr=&client=firefox-a&rls=org.mozilla%3Aen-US%3Aofficial&q=property+prediction+drug&btnG=Search > > > simple inspection of the literature - the majority of intermediates made > > during the synthesis of a new compound are themselves also new compounds. > > I agree that direct hits will be rare. Similiarities won't be so rare. > De novo predictions will almost always predict a property, and most > of them will be even not completely out of it. > > > The question of assessing synthetic feasibility arises from de novo design > > of candidate structures, and only after they have been found to be novel > > compounds. Obviously, if the computational chemist suggests a > > It would be too resource-intensive to check a structure for novelty > at generation time. Furthermore it would be pointless: nobody has > screened all commercially available and ever synthesized compounds against > a specific target, and made the results freely available. > > > commercially-available compound or one for which a synthesis has already > > been published, the "accessibility" of that compound is known and > > established. > > When stepping through structure and conformation space, enumerating or randomly > sampling cavity-constrained molecules the system does not know whether the matches > it generates have been ever made by somebody, somewhere. It would not make sense to > prune at that stage. Odds are overwhelming, anyway, that structure has never > been made -- but that a similiar structure almost certainly has been, and that > the synthetic route is in the literature, and that in some rare cases it even might be > available off-the shelf in sufficient quantities for a screen. > > The structure generator will easily drown you in structures it thinks are neat, > and ought to be screened. But in practice the guesses are not that good, and it > is not cost-effective to synthesize even a fraction of the lead candidates. > It would probably make more sense use that lead candidate as a starting point > for a combinatorial library, and hope that the diversity ball it covers will > contain real leads. As established by a honest, physical screen. > > > Even in this case, however, the properties of many of these substances are > > only incompletely catalogued in the literature. > > The only relevant property in the drug discovery: does it bind, or not, is > virtually guaranteed to not be on file. > > > Finally, as a synthetic chemist who dabbles in computational chemistry, I > > can assure everyone that literature precedent is not a guarantee of > > identical chemo-, regio- or stereoselectivity when the published reaction is > > applied to a new molecule. Yes, IF one is simply repeating the exact > > reaction reported in the paper, one should get the same result, but again, > > we are talking about de novo prediction on new chemical substances here. > > You may or may not get the same levels of selectivity when the reaction is > > applied to your compound. > > All good points. > > > The synthetic chemist can make educated guesses as to the breadth of > > applicability of any published method. The computer can also be programmed > > to do this, possibly using "expert system" methods etc. The question is how > > much effort is needed to make the prediction reliable? Isn't it simpler and > > probably cheaper to just use the expertise of the med chem. department in > > your company? > > Pharma industry is in a bind. The current rate of drug discovery does not > return on investment, and yet the trend is still getting worse. The current situation > is not sustainable, so both cost reductions and more accurate predictions > methods will be required. > > Highly trained specialists and animal facilities are an expensive and/or > scarce resource. There is no alternative to less people, and more models, > both in machina, in vitro and in vivo, and screening automation. The good > news is that the models never get worse. They only get better. > > -- > Eugen* Leitl leitl > ______________________________________________________________ > ICBM: 48.07100, 11.36820 http://www.leitl.org > 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE> > > From owner-chemistry@ccl.net Fri Sep 16 22:12:37 2005 From: "CCL" To: CCL Subject: CCL: W:ChemScore Message-Id: <-29217-050916165338-6849-4s151BQXxUrvrjgJG9iAzw*o*server.ccl.net> X-Original-From: "Robert Jorissen" Sent to CCL by: "Robert Jorissen" [jorissen*o*umbi.umd.edu] --Replace strange characters with the "at" sign to recover email address--. Hi I am currently implementing the ChemScore function of Eldridge et al., 1997 (JCAMD 11:425-445) for scoring docked conformations as generated by our in-house docking program, and would like to hear from other people who have implemented ChemScore. In particular, I would like to know what set of vdW radii are used; the radii are required for calculating the lipophilic contribution to this scoring function. Additionally, I would like to know of any example data that would allow me to assess whether I have implemented the individual terms (hydrogen bond, lipophilic, metal and rotatable bond terms) correctly. Any responses would be appreciated. Robert Jorissen Center for Advanced Research in Biotechnology