From owner-chemistry@ccl.net Sat Feb 16 05:40:01 2008 From: "Naser Eltaher Eltayeb nasertaha90()yahoo.co.uk" To: CCL Subject: CCL: chlorine parameters in ZINDO/S Message-Id: <-36310-080216053824-13254-2JdGlvhphhFByyZDIPJH4g{}server.ccl.net> X-Original-From: "Naser Eltaher Eltayeb" Date: Sat, 16 Feb 2008 05:38:21 -0500 Sent to CCL by: "Naser Eltaher Eltayeb" [nasertaha90:_:yahoo.co.uk] Dear All How can I add the paramters of chlorine to ZINDO/S in Gaussia 03? Thank you Naser From owner-chemistry@ccl.net Sat Feb 16 12:43:00 2008 From: "John J. Irwin jji .. cgl.ucsf.edu" To: CCL Subject: CCL: vHTS: DOCK or AUTODOCK ? Message-Id: <-36311-080215232753-25381-BZ3ZruyQLjACxJ1vDIoyRw*|*server.ccl.net> X-Original-From: "John J. Irwin" Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=ISO-8859-1; format=flowed Date: Fri, 15 Feb 2008 20:27:35 -0800 MIME-Version: 1.0 Sent to CCL by: "John J. Irwin" [jji!^!cgl.ucsf.edu] Dear Jerome You wrote > So my questions: do you know how DOCK & AUTODOCK handle vHTS of large > databases? Any personal experience or additional papers? Any one of > these two programs is better than the other for vHTS? I can tell you my personal experience with DOCK 3.5.54, a version of UCSF DOCK. This is our main workhorse, used for all the papers on our website. http://shoichetlab.compbio.ucsf.edu/publications.php. Throughput varies by the openness of the binding site, the actual speed of the CPU, the degree of sampling, size of ligands, and options selected. Typically, DOCK 3.5.54 processes 1 molecule in 0.1 to 10 seconds, with a typical value of 1-2 seconds. Let me multiply that out to make my point. We process 40-80K molecules per day per CPU. On our 500 CPU cluster, that lets us process around 20M-40M molecules per day. This may be biased somewhat by the fact that we favor "lead-like" or even "fragment-like" compounds over bigger molecules. Docking, in our experience, is more useful, and likely to be successful, with smaller molecules. The other caveat is that DOCK 3.5.54 is hard to use. That is why we created a web based interface, DOCK Blaster, at http://blaster.docking.org/. It goes both ways: I don't think we would have dared to create DOCK Blaster _unless_ DOCK 3.5.54 was fast. I have much less experience with DOCK 6, but believe the range of performance is both wider and has a substantially higher mean time per molecule. I believe we managed to screen DUD pretty well with as little as 14 seconds per molecule per CPU. (unpublished). The DOCK 5 paper quotes 5 min per ligand for saturated sampling flexible docking - in the abstract of the dock 5 paper. DOCK 6 is slightly faster, but perhaps not much. My general advice, if I dare, is to focus on results. Can you recapitulate what you already know? Can you convince yourself that you would have discovered what you know in a "blind" prospective trial, e.g. using a DUD-like approach. It really doesn't matter how fast or slow the program is - if these things aren't true, you need to try to make them true. We use DOCK 3.5.54 because we have convinced ourselves of this (see the DUD paper, Huang, Shoichet, Irwin JMC 2006). But it is a constant battle, and the universe is against you. Good luck! John http://johnirwin.compbio.ucsf.edu UCSF Pharmaceutical Chemistry and http://bluedolphinleads.com/ From owner-chemistry@ccl.net Sat Feb 16 13:18:00 2008 From: "John J Irwin jji- -cgl.ucsf.edu" To: CCL Subject: CCL: vHTS: DOCK or AUTODOCK ? Message-Id: <-36312-080216094149-21535-11RZJJerPUbYnNTGkiYBTw:-:server.ccl.net> X-Original-From: "John J Irwin" Date: Sat, 16 Feb 2008 09:41:43 -0500 Sent to CCL by: "John J Irwin" [jji-,-cgl.ucsf.edu] Dear Jerome You wrote > So my questions: do you know how DOCK & AUTODOCK handle vHTS of large databases? Any personal experience or additional papers? Any one of these two programs is better than the other for vHTS? I can tell you my personal experience with DOCK 3.5.54, a version of UCSF DOCK. This is our main workhorse, used for all the papers on our website. http://shoichetlab.compbio.ucsf.edu/publications.php. Throughput varies by the openness of the binding site, the actual speed of the CPU, the degree of sampling, size of ligands, and options selected. Typically, DOCK 3.5.54 processes 1 molecule in 0.1 to 10 seconds, with a typical value of 1-2 seconds. Let me multiply that out to make my point. We process 40-80K molecules per day per CPU. On our 500 CPU cluster, that lets us process around 20M-40M molecules per day. This may be biased somewhat by the fact that we favor "lead-like" or even "fragment-like" compounds over bigger molecules. Docking, in our experience, is more useful, and likely to be successful, with smaller molecules. The other caveat is that DOCK 3.5.54 is hard to use. That is why we created a web based interface, DOCK Blaster, at http://blaster.docking.org/. It goes both ways: I don't think we would have dared to create DOCK Blaster _unless_ DOCK 3.5.54 was fast. I have much less experience with DOCK 6, but believe the range of performance is both wider and has a substantially higher mean time per molecule. I believe we managed to screen DUD pretty well with as little as 14 seconds per molecule per CPU. (unpublished). The DOCK 5 paper quotes 5 min per ligand for saturated sampling flexible docking - in the abstract of the dock 5 paper. DOCK 6 is slightly faster, but perhaps not much. My general advice, if I dare, is to focus on results. Can you recapitulate what you already know? Can you convince yourself that you would have discovered what you know in a "blind" prospective trial, e.g. using a DUD-like approach. It really doesn't matter how fast or slow the program is - if these things aren't true, you need to try to make them true. We use DOCK 3.5.54 because we have convinced ourselves of this (see the DUD paper, Huang, Shoichet, Irwin JMC 2006). But it is a constant battle, and the universe is against you. Good luck! John http://johnirwin.compbio.ucsf.edu UCSF Pharmaceutical Chemistry and http://bluedolphinleads.com/ From owner-chemistry@ccl.net Sat Feb 16 13:54:01 2008 From: "John J Irwin jji ~ cgl.ucsf.edu" To: CCL Subject: CCL: vHTS: DOCK or AUTODOCK ? Message-Id: <-36313-080215234753-26395-S2Vr3CvhRXlKBoA0eyvlgQ__server.ccl.net> X-Original-From: "John J Irwin" Date: Fri, 15 Feb 2008 23:47:49 -0500 Sent to CCL by: "John J Irwin" [jji^^^cgl.ucsf.edu] Dear Jerome You wrote > So my questions: do you know how DOCK & AUTODOCK handle vHTS of large databases? Any personal experience or additional papers? Any one of these two programs is better than the other for vHTS? I can tell you my personal experience with DOCK 3.5.54, a version of UCSF DOCK. This is our main workhorse, used for all the papers on our website. http://shoichetlab.compbio.ucsf.edu/publications.php. Throughput varies by the openness of the binding site, the actual speed of the CPU, the degree of sampling, size of ligands, and options selected. Typically, DOCK 3.5.54 processes 1 molecule in 0.1 to 10 seconds, with a typical value of 1-2 seconds. Let me multiply that out to make my point. We process 40-80K molecules per day per CPU. On our 500 CPU cluster, that lets us process around 20M-40M molecules per day. This may be biased somewhat by the fact that we favor "lead-like" or even "fragment-like" compounds over bigger molecules. Docking, in our experience, is more useful, and likely to be successful, with smaller molecules. The other caveat is that DOCK 3.5.54 is hard to use. That is why we created a web based interface, DOCK Blaster, at http://blaster.docking.org/. It goes both ways: I don't think we would have dared to create DOCK Blaster _unless_ DOCK 3.5.54 was fast. I have much less experience with DOCK 6, but believe the range of performance is both wider and has a substantially higher mean time per molecule. I believe we managed to screen DUD pretty well with as little as 14 seconds per molecule per CPU. (unpublished). The DOCK 5 paper quotes 5 min per ligand for saturated sampling flexible docking - in the abstract of the dock 5 paper. DOCK 6 is slightly faster, but perhaps not much. My general advice, if I dare, is to focus on results. Can you recapitulate what you already know? Can you convince yourself that you would have discovered what you know in a "blind" prospective trial, e.g. using a DUD-like approach. It really doesn't matter how fast or slow the program is - if these things aren't true, you need to try to make them true. We use DOCK 3.5.54 because we have convinced ourselves of this (see the DUD paper, Huang, Shoichet, Irwin JMC 2006). But it is a constant battle, and the universe is against you. Good luck! John http://johnirwin.compbio.ucsf.edu UCSF Pharmaceutical Chemistry and http://bluedolphinleads.com/ From owner-chemistry@ccl.net Sat Feb 16 20:14:01 2008 From: "John McKelvey jmmckel,,gmail.com" To: CCL Subject: CCL: Computing on graphical processors Message-Id: <-36314-080216201204-9600-rDGYi00AEj7utOzUeR+uOQ{:}server.ccl.net> X-Original-From: "John McKelvey" Content-Type: multipart/alternative; boundary="----=_Part_3227_12864297.1203210714029" Date: Sat, 16 Feb 2008 20:11:54 -0500 MIME-Version: 1.0 Sent to CCL by: "John McKelvey" [jmmckel#gmail.com] ------=_Part_3227_12864297.1203210714029 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Hello.. It seems that there is beginning to be significant efforts to do processing using video card processors. Could someone comment on this as to why this is being so successful [probably many factors, I suppose], what kinds of computations work well on them, and what kinds of work probably do not work well on them.... and why? Thanks! John McKelvey ------=_Part_3227_12864297.1203210714029 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Hello..

It seems that there is beginning to be significant efforts to do processing using video card processors.   Could someone comment on this as to why this is being so successful [probably many factors, I suppose], what kinds of computations work well on them, and what kinds of work probably do not work well on them.... and why?

Thanks!

John McKelvey
------=_Part_3227_12864297.1203210714029-- From owner-chemistry@ccl.net Sat Feb 16 23:28:01 2008 From: "Alejandro Pisanty apisan]|[servidor.unam.mx" To: CCL Subject: CCL: Computing on graphical processors Message-Id: <-36315-080216223626-26945-A43uUxcmGxbtseM5yHeKIg * server.ccl.net> X-Original-From: Alejandro Pisanty Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed Date: Sun, 17 Feb 2008 02:37:01 +0000 (UTC) MIME-Version: 1.0 Sent to CCL by: Alejandro Pisanty [apisan]_[servidor.unam.mx] John, with great memories from first meeting you in 1979 in a now-historic workshop on computational chemistry methods, and again in 1991 for the meeting of which sprang this list - and in fortunate coincidence as I am just typing an entry in my blog, http://pisanty.blogspot.com (in Spanish) about the vitality of the CCL, a quick comment on your question: 1. some graphical processors are faster and more efficient, if properly programmed, than PC CPU's; 2. there are a lot of surplus powerful GPUs left after PCs and workstations are discarded or substituted with more powerful systems; 3. there is some literature reporting successful ports of useful codes (I'll defer to my colleague Carlos Amador, from Facultad de Quimica, UNAM, Mexico to provide exact references, among which a recent paper of his); 4. it would seem there are serious scaling problems to be dealt with for people who want to put together the power of many processors; building a cluster of GPUs is not without challenges. Yours, Alejandro Pisanty . . . . . . . . . . . . . . . . . . . . . . . . . . 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 *Mi blog/My blog: http://pisanty.blogspot.com *LinkedIn profile: http://www.linkedin.com/in/pisanty *Unete al grupo UNAM en LinkedIn, http://www.linkedin.com/e/gis/22285/4A106C0C8614 ---->> Unete a ISOC Mexico, www.isoc.org Participa en ICANN, www.icann.org . . . . . . . . . . . . . . . . . . . . . . . . . . On Sat, 16 Feb 2008, John McKelvey jmmckel,,gmail.com wrote: > Date: Sat, 16 Feb 2008 20:11:54 -0500 > From: "John McKelvey jmmckel,,gmail.com" > Reply-To: CCL Subscribers > To: "Pisanty, Alejandro " > Subject: CCL: Computing on graphical processors > > Hello.. > > It seems that there is beginning to be significant efforts to do processing > using video card processors. Could someone comment on this as to why this > is being so successful [probably many factors, I suppose], what kinds of > computations work well on them, and what kinds of work probably do not work > well on them.... and why? > > Thanks! > > John McKelvey >