From owner-chemistry@ccl.net Sat Nov 24 03:24:01 2007 From: "Ming Hui Yong yminghui=-=gmail.com" To: CCL Subject: CCL: Computational binding energy calculations Message-Id: <-35711-071123231650-9086-46TsBtbBTvsGzHr6xM2k9g!A!server.ccl.net> X-Original-From: "Ming Hui Yong" Date: Fri, 23 Nov 2007 23:16:47 -0500 Sent to CCL by: "Ming Hui Yong" [yminghui(_)gmail.com] Dear List, When computationally computing the binding energy for A + B --> AB, e.g. in AMBER's MMGBSA I calculate G(AB) - G(A) - G(B), for the case when 2 small ligands (A and B) bind to a protein P at different sites (I do not know the order however): say P + A --> PA --(1) PA + B --> PAB --(2). i)it could be that both P + A -->PA and P + B --> PB can occur in the presence of only A or B or ii)that B can only bind to P after A has bound to P what is the relevant computational binding energetics analyses/calculations to characterize the process for each case? I'm not sure what is done experimentally (e.g. using ITC). e.g. deltaG1 = PA - P - A and deltaG2 = PAB - PA - B or simply deltaG(overall) = PAB - P - A - B (would this be correct)?; and if say if I should calculate all 3, and the deltaG(overall) calculated is not equal to the sum of deltaG1 and deltaG2, can I still interpret the ratio of deltaG1 to deltaG2 as A and B's relative contributions to the overall binding energy? Comparing(PAB - A - PB) and (PAB - B - PA) would be relevant for studying cooperative binding, I suppose? Thank you all for your time. Regards, Ming Hui From owner-chemistry@ccl.net Sat Nov 24 04:41:01 2007 From: "chupvl chupvl[a]gmail.com" To: CCL Subject: CCL: statistical methods Message-Id: <-35712-071124043125-27300-E5tY7DZ828U7hiJNIq1vMQ[a]server.ccl.net> X-Original-From: chupvl Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=ISO-8859-1; format=flowed Date: Sat, 24 Nov 2007 11:33:34 +0300 MIME-Version: 1.0 Sent to CCL by: chupvl [chupvl[*]gmail.com] Hi! It's depends on training set used and purposes of the research and molecular descriptors you are going to use. Every method have it's own benefits and disadvantages. You'd better read some review papers about QSAR methods or some books. Vladimir Chupakhin ilknurca41#gmail.com wrote: > Sent to CCL by: ilknurca41*o*gmail.com > > Hi everybody, > What do you think about which is the most effective statistical > method in drug design using QSAR?(Multiple Lineer Regression, > Partial Least Square, Principle Component Analysis, Neural Networks, > etc....) > > Thanks for attention. Kind regards... > > ilknurca41:-:gmail.com> > > > > From owner-chemistry@ccl.net Sat Nov 24 18:37:01 2007 From: "Thomas P. Stockfisch t.stockfisch%a%cox.net" To: CCL Subject: CCL: statistical methods Message-Id: <-35713-071124175805-19184-bzQCqE8FTruU3rS6kyKiyQ^^^server.ccl.net> X-Original-From: "Thomas P. Stockfisch" Date: Sat, 24 Nov 2007 17:58:01 -0500 Sent to CCL by: "Thomas P. Stockfisch" [t.stockfisch(~)cox.net] Ilknurca41 - There is no one best statistical model building method for QSAR. It depends on your problem, the software you are using, and how much effort you are willing to expend fine-tuning and validating. Some methods handle only categorical output, some handle only numeric, and some handle both. Some can create non-linear mappings, some can't. Some are good for "black box" prediction, others are better for creating a simple, easy to interpret model that can be reverse-engineered into new candidate molecules. Some handle noisy data with many "hidden" variables, some are not so good for this. Some are tolerant of outliers, others require outliers to be identified and removed. Some methods are best used just for organizing and categorizing your data, but not for prediction. Some methods do everything for you and present you with one answer, others require massaging of input data, adjusting parameters, and validating chosen models. Some can accommodate new incoming training data and produce an updated model without access to the original data set. Some work well with thousands of input variables and dozens of output variables. Do you have a particular project in mind or are you looking to start a general discussion? Tom Stockfisch t.stockfisch ~ cox.net http://www.tstockfisch.com >On Nov 20, 2007, at 10:51 PM, ilknurca41#gmail.com wrote: Sent to CCL by: ilknurca41*o*gmail.com >Hi everybody, >What do you think about which is the most effective statistical >method in drug design using QSAR?(Multiple Lineer Regression, >Partial Least Square, Principle Component Analysis, Neural Networks, >etc....) >Thanks for attention. Kind regards... >ilknurca41:-:gmail.com From owner-chemistry@ccl.net Sat Nov 24 22:51:01 2007 From: "Randall H Goldsmith CCLrandall|*|yahoo.com" To: CCL Subject: CCL:G: NBO analysis followed by new SCF cycle Message-Id: <-35714-071124222259-9851-e5ohgswvS9bgfK0Plpc8jw..server.ccl.net> X-Original-From: "Randall H Goldsmith" Date: Sat, 24 Nov 2007 22:22:55 -0500 Sent to CCL by: "Randall H Goldsmith" [CCLrandall**yahoo.com] hello, While running Gaussian 98/NBO 3.1, I am attempting to do a deletion analysis, but instead of just looking at the new SCF energy, I want to look at the new molecular orbitals (not the new NBO's) that were created from the deletions in the fock matrix that I inputed. It would be most convenient to be able to generate a formatted checkpoint file of these new MO's. thanks! randall.