From chemistry-request@server.ccl.net Thu Apr 4 02:00:22 2002 Received: from haney.hbond.com ([207.137.2.252]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g3470Kp12820 for ; Thu, 4 Apr 2002 02:00:20 -0500 Received: (from haney@localhost) by haney.hbond.com (8.9.3/8.9.3) id XAA03270; Wed, 3 Apr 2002 23:02:04 -0800 From: "Dr. David N. Haney" Message-Id: <200204040702.XAA03270@haney.hbond.com> Subject: Bioinformatics Software Tools To: CHEMISTRY@ccl.net Date: Wed, 3 Apr 2002 23:02:04 -0800 (PST) Cc: haney@haney.hbond.com (Dr. David N. Haney) X-Mailer: ELM [version 2.5 PL5] MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit eduSoft, LC Announces New Bioinformatics Software: ### Molconn-Z DEVELOPERS' TOOLKIT ### for UNIX/LINUX and Windows Molconn-Z is now available as a Developers TOOLKIT. -SGI IRIX, Sun Solaris, Linux, Windows 2000 -over 300 MOLECULAR DESCRIPTORS: the standard for DESCRIPTOR analysis -integrate with in-house tools or re-package for clients -DATABASE + MOLCONN-Z + STATISTICS --> New Leads ### Molconn-Z JAVA Version ### for UNIX/LINUX and Windows Molconn-Z is now available with a Java Interface. -SGI IRIX, Sun Solaris, Linux, Windows 2000. Beta testers needed. -Statistical Analysis in Java Tables -Client-Server methodology to allow JAVA on 1 machine and Molconn-Z elsewhere ### HINT DEVELOPERS' TOOLKIT ### for UNIX/LINUX and Windows -LogP for small molecules, macromolecules, 3D LogP maps, dG interaction maps -3D LogP QSAR fields, 3D LogP docking score, 3D molecule interaction maps -integrate with in-house tools or re-package for clients website http://www.edusoft-lc.com. Software is available for free trial. VISIT eduSoft and OpenEye Software at the ACS: April 8-10 Booths #170,172! -- ######### David N. Haney, Ph.D. ######### # Haney Associates Phone - 858-483-1197 # # 5455 Westknoll Dr. FAX - 858-483-1046 # # La Jolla, CA 92037 Email - haney@hbond.com # ################# ##################### From chemistry-request@server.ccl.net Thu Apr 4 05:30:29 2002 Received: from biochim.ro ([193.231.158.224]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34ATqp18842 for ; Thu, 4 Apr 2002 05:29:59 -0500 Received: from localhost (amilac@localhost) by biochim.ro (8.9.3/8.9.3) with ESMTP id NAA11035 for ; Thu, 4 Apr 2002 13:40:48 -0200 Date: Thu, 4 Apr 2002 13:40:48 -0200 (GMT+2) From: Adina Milac To: Computational Chemistry List Subject: neural networks problem Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII We are trying to develop a neural network that would be able to predict the inhibition of some chemical compounds based on some physico-chemical properties of these compounds. From the mathematical point of view, the problem is to associate training vectors with 4 fields with a scalar. The problem we face is the following: after we have tried different variants of training sets, we have found that if "extreme" vectors (vectors in which one of the fields has either maximum or minimum values) are excluded from the training set, the prediction for that vector is extremely poor. In other words, the network is not able to extrapolate the data it learns. In my opinion, this could be due to one (or more) of the following reasons (from least to most probable): 1. network architecture is not appropriate for the task (although I have tried many variants of network dimensions and transfer functions) 2. the network is unable to determine a function based on which it could make the extrapolation 3. the physico-chemical properties we have used for building input vectors are not adequate for predicting inhibition What could be, in your opinion, the cause for this failure and what do you suggest me to do? Hoping that you will find the time to answer my message, I thank you and I wish you all the best, ADINA MILAC Ph.D Student, Research Assistant Institute of Biochemistry, Romanian Academy Splaiul Independentei 296, 77700 Bucharest 17, Romania Phone: (+401).223.90.69; FAX: (+401).223.90.68 e-mail: amilac@biochim.ro From chemistry-request@server.ccl.net Thu Apr 4 03:41:01 2002 Received: from mh1-cam.cam.accelrys.com ([192.190.183.253]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g348f0p15294 for ; Thu, 4 Apr 2002 03:41:01 -0500 Received: from camdmail01.cam.accelrys.com ([172.27.160.90]) by mh1-cam.cam.accelrys.com with esmtp (Exim 3.12 #1 (Debian)) id 16t2nW-0001z8-00 for ; Thu, 04 Apr 2002 09:40:46 +0100 To: "chemistry@ccl.net" Subject: Accelrys Life Science Training in Paris, France MIME-Version: 1.0 X-Mailer: Lotus Notes Release 5.0.9a January 7, 2002 Message-ID: From: tluu@accelrys.com Date: Thu, 4 Apr 2002 09:40:43 +0000 X-MIMETrack: Serialize by Router on camdmail01/Server/Accelrys(Release 5.0.9 |November 16, 2001) at 04/04/2002 09:40:46 AM, Serialize complete at 04/04/2002 09:40:46 AM Content-Type: multipart/alternative; boundary="=_alternative 002FAC9E80256B91_=" This is a multipart message in MIME format. --=_alternative 002FAC9E80256B91_= Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Accelrys will be holding a pair of 2-day workshops at the Universit=E9=20 Pierre et Marie Curie, Paris, France. On May 21-22, the "Introduction to Cerius2 for Life Sciences" workshop=20 will be offered. This course provides an overview of molecular modelling=20 techniques for life sciences applications using Cerius2. The workshop=20 will focus on basic skills for use of the interface and will explore=20 various modules in the areas of rational-drug design, structure-based drug = design, and combinatorial chemistry. Prior modelling experience is not=20 assumed making this course a great place to learn molecular modelling with = Cerius2. On May 23-24, the " Small Molecule and Drug Design with Cerius2" workshop=20 will be offered. This workshop is aimed at our customers who are involved = in drug design or the development of other bioactive compounds and who=20 would like to make more effective use of modelling in their research. The = course will focus on QSAR techniques and methodologies as well as=20 structure-based ligand design. Attendees should possess knowledge of=20 basic UNIX commands and have a basic understanding of QSAR theory.=20 Familiarity with the Cerius2 environment is required for attendance to=20 this workshop. Both suggested prerequisites can be met by attending the=20 introductory workshop. Fees for each 2-day course are 1200 euro commercial, 900 euro government,=20 and 660 euro academic. If payment is made through a credit card, the fees = are US$1200 commercial, US$900 government, and US$660 academic (excl VAT). = However, register for both courses and receive a 10% discount for the=20 second course. Registration is on-line at URL http://www.accelrys.com/training/lifesci/reg= istration.php. Further detailed information about this and other Accelrys = training=20 workshops can be found at the Accelrys website (http://www.accelrys.com/tra= ining/lifesci/schedule.html). Please do not hesitate to contact us should = you have any questions. Thank you very much. Tien Luu +44 1223 402 895 Judith Madeley +44 1132 033 714 --=_alternative 002FAC9E80256B91_= Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable
Accelrys will be holding a pair of 2= -day workshops at the Universit=E9 Pierre et Marie Curie, Paris, France.

On May 21-22, the "Introduction= to Cerius2 for Life Sciences" workshop will be offered.  This co= urse provides an overview of molecular modelling techniques for life scienc= es applications using Cerius2.  The workshop will focus on basic skill= s for use of the interface and will explore various modules in the areas of= rational-drug design, structure-based drug design, and combinatorial chemi= stry.  Prior modelling experience is not assumed making this course a = great place to learn molecular modelling with Cerius2.

On May 23-24, the " Small Molec= ule and Drug Design with Cerius2" workshop will be offered.  This= workshop is aimed at our customers who are involved in drug design or the = development of other bioactive compounds and who would like to make more ef= fective use of modelling in their research.  The course will focus on = QSAR techniques and methodologies as well as structure-based ligand design.=  Attendees should possess knowledge of basic UNIX commands and have a= basic understanding of QSAR theory.  Familiarity with the Cerius2 env= ironment is required for attendance to this workshop.  Both suggested = prerequisites can be met by attending the introductory workshop.

Fees for each 2-day course are 1200 = euro commercial, 900 euro government, and 660 euro academic.  If payme= nt is made through a credit card, the fees are US$1200 commercial, US$900 g= overnment, and US$660 academic (excl VAT).  However, register for both= courses and receive a 10% discount for the second course.

Registration is on-line at URL http:= //www.accelrys.com/training/lifesci/registration.php.  Further detaile= d information about this and other Accelrys training workshops can be found= at the Accelrys website (http://www.accelrys.com/training/lifesci/schedule= .html).  Please do not hesitate to contact us should you have any ques= tions.


Thank you very much.

Tien Luu
+44 1223 402 895

Judith Madeley
+44 1132 033 714 --=_alternative 002FAC9E80256B91_=-- From chemistry-request@server.ccl.net Thu Apr 4 08:45:02 2002 Received: from igc.phys.chem.ethz.ch ([129.132.218.130]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34Divp22477 for ; Thu, 4 Apr 2002 08:44:58 -0500 Received: from humagne.ethz.ch (humagne [129.132.218.162]) by igc.phys.chem.ethz.ch (8.11.4/8.11.4) with ESMTP id g34Dimx27817 for ; Thu, 4 Apr 2002 15:44:48 +0200 (MEST) Received: (from portmann@localhost) by humagne.ethz.ch (8.11.4/8.11.4) id g34Dimj11419 for chemistry@ccl.net; Thu, 4 Apr 2002 15:44:48 +0200 (MEST) Date: Thu, 4 Apr 2002 15:44:48 +0200 (MEST) Message-Id: <200204041344.g34Dimj11419@humagne.ethz.ch> From: Stefan Portmann Reply-To: info@watoc02.ch Subject: WATOC'02: First detailed program available Content-Type: text Dear Madam or Sir 6th WORLD CONGRESS OF THEORETICALLY ORIENTED CHEMISTS (WATOC'02) August 4-9, 2002, Palazzo dei Congressi, Lugano, Switzerland Theory, Computation and Information Science in Chemistry, Biochemistry and Materials Science We are pleased to announce the FIRST DETAILED PROGRAM of WATOC'02 at http://www.watoc02.ch Please also have a look at the information regarding local expenses in the section "Budgetary Considerations". Don't miss the deadline to get the early bird discount (April 31 2002) and register also at http://www.watoc02.ch Abstracts for poster presentations can still be submitted at the same location. The poster sessions will be endowed by Gaussian, Inc.. In order to give the poster sessions optimal visibility, an award program was established. The posters will be rated both on popular and on jury votes. There will be a two hour award ceremony on Thursday evening. We are looking forward to see you in Lugano. Hans Peter Luethi and Stefan Portmann ETH Zurich From chemistry-request@server.ccl.net Thu Apr 4 07:17:22 2002 Received: from mail1.rrz.Uni-Koeln.DE ([134.95.100.208]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34CHMp20724 for ; Thu, 4 Apr 2002 07:17:22 -0500 Received: from campfire.rrz.Uni-Koeln.DE (IDENT:39358@campfire.rrz.Uni-Koeln.DE [134.95.19.27]) by mail1.rrz.Uni-Koeln.DE (8.12.2/8.12.2) with ESMTP id g34CHFmY029762 for ; Thu, 4 Apr 2002 14:17:15 +0200 (MEST) Received: (from a0580@localhost) by campfire.rrz.Uni-Koeln.DE (8.9.1a/8.9.1) id OAA17653 for chemistry@ccl.net; Thu, 4 Apr 2002 14:17:13 +0200 (MET DST) Date: Thu, 4 Apr 2002 14:17:13 +0200 From: Lars Packschies To: CCL Subject: G98 - Arithmetic Exception ? Message-ID: <20020404121713.GC2307@campfire.rrz.Uni-Koeln.DE> Mail-Followup-To: CCL Mime-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="p2kqVDKq5asng8Dg" Content-Disposition: inline User-Agent: Mutt/1.3.28i X-Virus-Scanned: by amavisd-milter (http://amavis.org/) --p2kqVDKq5asng8Dg Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable Dear CCLers, has anyone ever seen this g98a9 (and g98a11) error message: Arithmetic Exception ? job: # opt rmp2/CEP-121G(3df) ..... adding "3df" there leads to the error... It's a sun fire 15000, running Solaris 8 G98: (`file` output)=20 g98: ELF 64-bit MSB executable, SPARC V9, version 1 (SYSV),=20 dynamically linked (uses shared libs), not stripped Greetings Lars --=20 ,H + Dr. Lars Packschies ZAIK/RRZK, Robert-Koch-Str.10 + H H O | Chemistry Dpt. Support D-50931 Cologne | `O-H. | ,' `H | Comp.Dpt., Univ. of Cologne Phn (+49)221-478-7022 | `O-H + Packschies@rrz.uni-koeln.de Fax (+49)221-478-5568 + --p2kqVDKq5asng8Dg Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.6 (SunOS) Comment: For info see http://www.gnupg.org iD8DBQE8rERHkRMt176/QQkRAit2AJ9qClW54Lsvx23bukKMM4w86o1HNgCgmOm2 kckXthMdLFLngpdLYqgDoKU= =6yM3 -----END PGP SIGNATURE----- --p2kqVDKq5asng8Dg-- From chemistry-request@server.ccl.net Wed Apr 3 16:51:36 2002 Received: from mail.rpi.edu ([128.113.22.40]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g33Lpap00461 for ; Wed, 3 Apr 2002 16:51:36 -0500 Received: from minghu (exothermic-40.dynamic.rpi.edu [128.113.21.193]) by mail.rpi.edu (8.12.1/8.12.1) with SMTP id g33LpNsU263818; Wed, 3 Apr 2002 16:51:23 -0500 Message-ID: <000a01c1db58$932ba2c0$c1157180@dynamic.rpi.edu> From: "Jerry Song" To: "Zhangxd" Cc: References: <003601c1db2d$ea779000$5dcae88c@pak3> Subject: Re: CCL:how i can get detailed secondary structure from coordinate file? Date: Wed, 3 Apr 2002 16:43:20 -0500 MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 5.00.2615.200 X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2615.200 X-Scanned-By: MIMEDefang 2.3 (www dot roaringpenguin dot com slash mimedefang) Rasmol. you can open it and then concert it to molscript file (you need to download the executable molscript program and set up the proper path), which will contain the secondry structure infomation you need. the format will be : ..... turn residue 10; turn residue 11; turn residue 12; set planecolour green; coil from 10 to 13; set planecolour rgb 1 0 0; helix from 13 to 28; turn residue 28; ...... > ----- Original Message ----- > From: "Zhangxd" > To: "Jeff Nauss" > Cc: > Sent: Monday, February 04, 2002 5:56 PM > Subject: CCL:how i can get detailed secondary structure from coordinate > file? > > > > Dear All, > > during my research, i need the detailed information about secondary > > structure of my biology system, i had its coordinate file, how i can get > > its secondary structure information from this coordinate file? just like > > the following: > > Resi. 1 to Resi. 10 AlphaHelix > > Resi. 15 to Resi. 25 BetaSheet > > >.................. > > >............... > > > > hopefully it should be a program can do this automatically. > > i had tried VMD, it just gives the graph of secondary structure of > > the sequence but not a detailed text information like the above sample. > > > > thanks in advance!!!! > > > > > > > > CHEMISTRY@ccl.net -- To Everybody | CHEMISTRY-REQUEST@ccl.net -- To > Admins > > CHEMISTRY-SEARCH@ccl.net -- archive search | Gopher: gopher.ccl.net > 70 > > Ftp: ftp.ccl.net | WWW: http://www.ccl.net/chemistry/ | Jan: > jkl@ccl.net > > > > > > > > > > > > > > -= This is automatically added to each message by mailing script =- > CHEMISTRY@ccl.net -- To Everybody | CHEMISTRY-REQUEST@ccl.net -- To Admins > MAILSERV@ccl.net -- HELP CHEMISTRY or HELP SEARCH > CHEMISTRY-SEARCH@ccl.net -- archive search | Gopher: gopher.ccl.net 70 > Ftp: ftp.ccl.net | WWW: http://www.ccl.net/chemistry/ | Jan: jkl@ccl.net > > > > > From chemistry-request@server.ccl.net Thu Apr 4 08:51:27 2002 Received: from uscmail.usc.es ([193.144.75.8]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34DpRp22546 for ; Thu, 4 Apr 2002 08:51:27 -0500 Received: from qfcronos (qfcronos.usc.es [193.144.74.154]) by uscmail.usc.es (8.9.1/8.9.1) with SMTP id PAA16830 for ; Thu, 4 Apr 2002 15:51:20 +0200 (MET DST) Message-ID: <003e01c1dbdf$e716b6e0$9a4a90c1@usc.es> From: =?iso-8859-1?Q?Jes=FAs_Rodr=EDguez_Otero?= To: Subject: anisotropy of the susceptibility with G98 Date: Thu, 4 Apr 2002 15:52:03 +0200 MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="----=_NextPart_000_003B_01C1DBF0.AA658B20" X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 6.00.2600.0000 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000 This is a multi-part message in MIME format. ------=_NextPart_000_003B_01C1DBF0.AA658B20 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Hi netters Does Gaussian calculate correctly the anisotropy of the magnetic = susceptibility? It seems to me that no. ********************************************************** Magnetic susceptibility (cgs-ppm): Isotropic =3D -91.9555 Anisotropy=3D 47.0081 XX=3D -61.8277 YX=3D -2.8296 ZX=3D 0.0070 XY=3D -2.7275 YY=3D -66.9920 ZY=3D 0.0003 XZ=3D -0.0009 YZ=3D 0.0198 ZZ=3D -147.0468 Eigenvalues: -147.0468 -68.2029 -60.6168 ********************************************************************* the correct thing is to calculate it of the following way? anisotropy=3DTzz-1/2(Txx+Tyy). This way the result is -82.64 or with the Eigenvalues?? This way the result is almost equal Thanks ------=_NextPart_000_003B_01C1DBF0.AA658B20 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable
Hi netters
 
Does Gaussian calculate correctly the anisotropy = of the=20 magnetic susceptibility?
It seems to me that no.
 
**********************************************************
Magnetic susceptibility = (cgs-ppm):
  =20 Isotropic =3D   -91.9555   = Anisotropy=3D   =20 47.0081
   XX=3D   -61.8277  =20 YX=3D    -2.8296   = ZX=3D    =20 0.0070
   XY=3D    -2.7275   = YY=3D  =20 -66.9920   ZY=3D     = 0.0003
  =20 XZ=3D    -0.0009   = YZ=3D    =20 0.0198   ZZ=3D  -147.0468
   = Eigenvalues: =20 -147.0468   -68.2029   -60.6168
****************************************************************= *****
 
the correct thing is to calculate it of the = following=20 way?
 
anisotropy=3DTzz-1/2(Txx+Tyy). This way the = result is=20 -82.64
 
or with the Eigenvalues?? This way the result is = almost=20 equal
 
Thanks
 
------=_NextPart_000_003B_01C1DBF0.AA658B20-- From chemistry-request@server.ccl.net Wed Apr 3 14:44:29 2002 Received: from research.dfci.harvard.edu ([155.52.50.28]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g33JiTp28304 for ; Wed, 3 Apr 2002 14:44:29 -0500 Received: from [155.52.22.78] (account reche HELO research.dfci.harvard.edu) by research.dfci.harvard.edu (CommuniGate Pro SMTP 3.5b5) with ESMTP-TLS id 641636 for chemistry@ccl.net; Wed, 03 Apr 2002 14:47:02 -0500 Sender: par@ccl.net Message-ID: <3CAB59AF.FB77CA3E@research.dfci.harvard.edu> Date: Wed, 03 Apr 2002 14:36:15 -0500 From: Pedro A Reche Gallardo Reply-To: reche@research.dfci.harvard.edu Organization: Dana-Farber Cancer Institute X-Mailer: Mozilla 4.7C-SGI [en] (X11; I; IRIX64 6.5 IP28) X-Accept-Language: en MIME-Version: 1.0 To: chemistry@ccl.net Subject: software molecular dymanics Content-Type: multipart/alternative; boundary="------------1ACF2B91AB0B4BAC8B2E23F4" --------------1ACF2B91AB0B4BAC8B2E23F4 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Dear All, I am interested in doing a molecular dynamics simulation on a ligand and the surrounding binding site in a protein keeping the rest fixed. Moreover, since I am planning to do this for a collection of ligands and proteins I would like to do it in an automated way based on running scripts. Unfortunately, I do not have much experience doing molecular dynamics, and therefore I will appreciate a lot if someone would let me know what the best software to use, and how to fix a set of atoms. Regards, Pedro Reche ******************************************************************* PEDRO A. RECHE , pHD TL: 617 632 3824 Dana-Farber Cancer Institute, FX: 617 632 4569 Harvard Medical School, EM: reche@research.dfci.harvard.edu 44 Binney Street, D1510A, EM: reche@mifoundation.org Boston, MA 02115 URL: http://www.reche.org ******************************************************************* --------------1ACF2B91AB0B4BAC8B2E23F4 Content-Type: text/html; charset=us-ascii Content-Transfer-Encoding: 7bit Dear All, I am interested in doing a molecular dynamics simulation on a ligand and the surrounding binding site in a protein keeping the rest fixed.  Moreover, since I am planning to do this for a collection of ligands and proteins I would like to do it in an automated way based on running scripts.  Unfortunately, I do not have much experience doing molecular dynamics, and  therefore I will appreciate a lot if someone would let me know what  the best software to use, and how to fix a set of atoms.
Regards,

Pedro Reche
*******************************************************************
PEDRO A. RECHE , pHD            TL: 617 632 3824
Dana-Farber Cancer Institute,   FX: 617 632 4569
Harvard Medical School,         EM: reche@research.dfci.harvard.edu
44 Binney Street, D1510A,       EM: reche@mifoundation.org              
Boston, MA 02115                URL: http://www.reche.org                                               
*******************************************************************
  --------------1ACF2B91AB0B4BAC8B2E23F4-- From chemistry-request@server.ccl.net Thu Apr 4 11:20:53 2002 Received: from ribotargets.com ([194.129.39.11]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34GKrp27649 for ; Thu, 4 Apr 2002 11:20:53 -0500 Received: from szilva (helo=localhost) by echo.ribotargets.com with local-esmtp (Exim 3.13 #1) id 16t9vS-0006Pf-00; Thu, 04 Apr 2002 16:17:26 +0000 Date: Thu, 4 Apr 2002 16:17:26 +0000 (GMT) From: Szilveszter Juhos To: Adina Milac cc: Computational Chemistry List Subject: Re: CCL:neural networks problem In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII On Thu, 4 Apr 2002, Adina Milac wrote: > In my opinion, this could be due to one (or more) of the following > reasons (from least to most probable): [..] Unfortunatelly neural nets are not that easy as some people think. In most of the cases the problem is what you mentioned: if you want to extrapolate to a region of parameter space that is not represeneted by the training data than you can have incorrect predictions. If I can recommend one book is that one that was written by C. Bishop ( http://www.ncrg.aston.ac.uk/People/bishopc/Welcome.html ) where most of these problems are discussed. Also there are a loads of good papers on these pages. My first run would be to play with the properties (3rd problem) checking the weights. If one of the properties have a very small weight than it is unlikely to affect the overall prediction so it can be replaced by an other property (that can improve prediction if we are lucky). Still likely you have to include extreme values into the training set. Szilva From chemistry-request@server.ccl.net Thu Apr 4 09:02:26 2002 Received: from shardagate.mahindrabt.com ([206.102.1.162]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34E2Pp23037 for ; Thu, 4 Apr 2002 09:02:25 -0500 Received: from mahindrabt.com (mailscan.sharda.mahindrabt.com [10.5.0.97]) by shardagate.mahindrabt.com (8.9.3/8.9.3/SuSE Linux 8.9.3-0.1) with ESMTP id TAA20025 for ; Thu, 4 Apr 2002 19:32:16 +0530 X-SMTP-Sending-IP: 10.5.0.15 Received: from intranet.sharda.mahindrabt.com by mahindrabt.com ; 4 Apr 2002 19:29:11 +0530 Received: from vijayp.mahindrabt.com ([10.5.2.102]) by intranet.sharda.mahindrabt.com (8.9.3/8.9.3) with SMTP id TAA29501; Thu, 4 Apr 2002 19:32:08 +0530 Message-ID: <002801c1dbe1$9f75b320$6602050a@mahindrabt.com> From: "Vijay Pargaonkar" To: "Adina Milac" , "Computational Chemistry List" References: Subject: Re: CCL:neural networks problem Date: Thu, 4 Apr 2002 19:34:22 +0530 MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 5.50.4522.1200 X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4522.1200 Dear Adina Milac, Your observations are right. I have also experienced the same fact about different data sets. As the degree of extrapolation increases the prediction capabilities of neural netwroks decreases. Neural Networks are efficient interpolators. They are not meant for extrapolations. Hence, Maxima and minima play crucial role in training set. There are reasons for poorer performance. I have a lot of experience on Neural Networks. However, Accurate information about your Neural Network will enable me in explaining you how to tackle this situation. Thanks. Pargaonkar Vijay ----- Original Message ----- From: "Adina Milac" To: "Computational Chemistry List" Sent: Thursday, April 04, 2002 9:10 PM Subject: CCL:neural networks problem > > We are trying to develop a neural network that would be able to predict > the inhibition of some chemical compounds based on some physico-chemical > properties of these compounds. From the mathematical point of view, the > problem is to associate training vectors with 4 fields with a scalar. > > The problem we face is the following: after we have tried different > variants of training sets, we have found that if "extreme" vectors > (vectors in which one of the fields has either maximum or minimum > values) are excluded from the training set, the prediction for that vector > is extremely poor. In other words, the network is not able to extrapolate > the data it learns. > > In my opinion, this could be due to one (or more) of the following > reasons (from least to most probable): > > 1. network architecture is not appropriate for the task (although I have > tried many variants of network dimensions and transfer functions) > > 2. the network is unable to determine a function based on which it could > make the extrapolation > > 3. the physico-chemical properties we have used for building input > vectors are not adequate for predicting inhibition > > What could be, in your opinion, the cause for this failure and what do you > suggest me to do? > > Hoping that you will find the time to answer my message, I thank you and I > wish you all the best, > > ADINA MILAC > > Ph.D Student, Research Assistant > Institute of Biochemistry, Romanian Academy > Splaiul Independentei 296, 77700 Bucharest 17, Romania > Phone: (+401).223.90.69; FAX: (+401).223.90.68 > e-mail: amilac@biochim.ro > > > -= This is automatically added to each message by mailing script =- > CHEMISTRY@ccl.net -- To Everybody | CHEMISTRY-REQUEST@ccl.net -- To Admins > MAILSERV@ccl.net -- HELP CHEMISTRY or HELP SEARCH > CHEMISTRY-SEARCH@ccl.net -- archive search | Gopher: gopher.ccl.net 70 > Ftp: ftp.ccl.net | WWW: http://www.ccl.net/chemistry/ | Jan: jkl@ccl.net > > > > > ********************************************************* Disclaimer This message (including any attachments) contains confidential information intended for a specific individual and purpose, and is protected by law. If you are not the intended recipient, you should delete this message and are hereby notified that any disclosure, copying, or distribution of this message, or the taking of any action based on it, is strictly prohibited. ********************************************************* Visit us at http://www.mahindrabt.com From chemistry-request@server.ccl.net Thu Apr 4 14:31:02 2002 Received: from is1.utmb.edu ([129.109.195.2]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34JV1p31511 for ; Thu, 4 Apr 2002 14:31:02 -0500 Received: from exchange3.utmb.edu (email.utmb.edu [129.109.52.245]) by utmb.edu (PMDF V6.1 #38124) with ESMTP id <0GU200DDW65X8R@utmb.edu> for chemistry@ccl.net; Thu, 04 Apr 2002 13:29:57 -0600 (CST) Received: by exchange3 with Internet Mail Service (5.5.2653.19) id ; Thu, 04 Apr 2002 13:30:54 -0600 Content-return: allowed Date: Thu, 04 Apr 2002 13:30:53 -0600 From: "Ivanciuc, Ovidiu I." Subject: RE: neural networks problem To: "'Adina Milac '" Cc: "'chemistry@ccl.net'" Message-id: <9BAF6985AAD2D4118DD100D0B79E84F104789B5C@exchange4> MIME-version: 1.0 X-Mailer: Internet Mail Service (5.5.2653.19) Content-type: text/plain; charset=iso-8859-1 Adina, >We are trying to develop a neural network that would be able to predict >the inhibition of some chemical compounds based on some physico-chemical >properties of these compounds. >The problem we face is the following: after we have tried different >variants of training sets, we have found that if "extreme" vectors >(vectors in which one of the fields has either maximum or minimum >values) are excluded from the training set, the prediction for that >vector is extremely poor. In other words, the network is not able to >extrapolate the data it learns. When using artificial neural networks (ANN) in QSAR you have to consider that ANN are highly non-linear systems in which extrapolation is very difficult. Moreover, the usual transfer function of the output neuron is the sigmoid which takes values between 0 and 1, or the hyperbolic tangent tanh which takes values between -1 and 1. Now suppose that in your calibration set the maximum output (Y exp value) is 4, and in the prediction set you have a pattern with the output 8. Now, in calibrating the ANN (optimization of the connection weights between neurons) the Yexp=4 is mapped to +1, which is the maximum output both for the sigmoid and tanh, and all weights are optimized to give an output value +1 for the input values associated with Yexp=4. When you test the pattern with the output value Yexp=8, your output will still be +1, since the sigmoid and tanh cannot give a larger output, irrespective of the input values. Therefore, you must understand that his is a mathematical limitation of ANN with sigmoid and tanh output transfer function. What can be done: - scale the output values in the calibration set of molecules to a smaller range, i.e. for tanh instead of -1 to +1 scale to -0.9 to +0.9. In this way, the ANN can offer an output value larger than the maximum output in the calibration set. - use a linear output transfer function, which is not bounded. In my QSAR and QSPR experiments, I have constantly found that a linear output function gives slightly smaller R in calibration but better R in prediction, compared with the sigmoid and tanh. - be aware that the ratio between the number of patterns used in calibration (to optimize the weights) and the number of optimizable parameters (number of connections between neurons) must be as large as possible, but not smaller than 5 or 4. - you have to optimize the weights with a good nonlinear optimization algorithm; the backpropagation is bad, and usually gives a cheap-and-dirty result. Go with a conjugate gradient algorithm. - always keep in mind that with QSAR models you cannot obtain good predictions for patterns very different from those used to calibrate the model. Succes si o zi buna, Ovidiu Ivanciuc ************************* Dr. Ovidiu Ivanciuc Sealy Center for Structural Biology, Department of Human Biological Chemistry & Genetics, University of Texas Medical Branch, 301 University Boulevard, Galveston, Texas 77555-1157 From chemistry-request@server.ccl.net Thu Apr 4 09:04:21 2002 Received: from unq.edu.ar ([200.49.110.238]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34E4Kp23087 for ; Thu, 4 Apr 2002 09:04:21 -0500 Received: from pII400Win95.unq.edu.ar [200.49.110.145] by unq.edu.ar [200.49.110.238] with SMTP (MDaemon.PRO.v5.0.1.R) for ; Thu, 04 Apr 2002 10:54:56 -0300 Message-Id: <5.1.0.14.0.20020404105414.00aaac00@correo.unq.edu.ar> X-Sender: mlema@correo.unq.edu.ar X-Mailer: QUALCOMM Windows Eudora Version 5.1 Date: Thu, 04 Apr 2002 11:04:04 -0300 To: chemistry@ccl.net From: =?iso-8859-1?Q?Mart=EDn?= Lema Subject: mutants with altered structures Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; format=flowed X-MDRemoteIP: 200.49.110.145 X-Return-Path: mlema@unq.edu.ar X-MDaemon-Deliver-To: chemistry@ccl.net Hello everybody: We are looking for examples of protein mutants (with only one or two mutations) whose crystallographic structures have been determined experimentally, and exhibit a significantly altered structure (different backbone trace, gyration radius, local secondary structure) when compared to their wild type counterparts. Any hint is welcome. Thanking in advance Martin Lema National University of Quilmes Argentina From chemistry-request@server.ccl.net Thu Apr 4 14:34:09 2002 Received: from ibitipoca.fisica.ufjf.br ([200.17.69.110]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34JY8p31542 for ; Thu, 4 Apr 2002 14:34:09 -0500 Received: from fisica.ufjf.br (ibitipoca.fisica.ufjf.br [200.17.69.110]) by ibitipoca.fisica.ufjf.br (8.11.2/8.11.2) with ESMTP id g34Jdqt17740; Thu, 4 Apr 2002 16:39:55 -0300 Sender: dantas@fisica.ufjf.br Message-ID: <3CACAC07.DE1C7AA6@fisica.ufjf.br> Date: Thu, 04 Apr 2002 16:39:51 -0300 From: Socrates de Oliveira Dantas Organization: Universidade Federal de Juiz de Fora X-Mailer: Mozilla 4.76 [en] (X11; U; Linux 2.4.2-2 i586) X-Accept-Language: en MIME-Version: 1.0 To: chemistry@ccl.net, dantas@fisica.ufjf.br Subject: Dear netters, Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit Dear netters, We need an advice on how to create a movie from xyz files (generated by moledular dynamics simulation). We have a thousand xyz files. Is there a script or another automated tool could help us to generate an AVI movie suitable to our needs? Could you help us with this problem? (an open or freeware software would be better) Thanks in advance. Sócrates Dantas. dantas@fisica.ufjf.br From chemistry-request@server.ccl.net Thu Apr 4 12:04:31 2002 Received: from rbadb2.rbacpxclu.bas.roche.com ([196.3.50.241]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g34H4Up28798 for ; Thu, 4 Apr 2002 12:04:31 -0500 Received: from CONVERSION-DAEMON.Roche.COM by Roche.COM (PMDF V6.0-025 #47170) id <01KG6CW8IC1C9AULMK@Roche.COM> for chemistry@ccl.net; Thu, 04 Apr 2002 19:01:47 +0200 Received: from rbamsemcn1.bas.roche.com (rbamsemcn1.bas.roche.com [145.245.211.139]) by Roche.COM (PMDF V6.0-025 #47170) with ESMTP id <01KG6CW64AHE9D4QOC@Roche.COM>; Thu, 04 Apr 2002 19:01:43 +0200 Received: by rbamsemcn1.bas.roche.com with Internet Mail Service (5.5.2653.19) id ; Thu, 04 Apr 2002 19:03:10 +0200 Content-return: allowed Date: Thu, 04 Apr 2002 19:02:58 +0200 From: "Borosy, Andras {Basi~Basel}" Subject: RE: neural networks problem To: "'Adina Milac'" , Computational Chemistry List Message-id: <5FBD82D1CB46D511AAEA0002A55C6F5203F168D8@rbamsem3.bas.roche.com> MIME-version: 1.0 X-Mailer: Internet Mail Service (5.5.2653.19) Content-type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from quoted-printable to 8bit by server.ccl.net id g34H4Vp28799 > > 2. the network is unable to determine a function based on > which it could > make the extrapolation > That is the main reason. However the other regression methods suffer from that, too. If one is lucky linear relationship may valid outside the training set, if the linearity is more general. Best wishes, András Borosy From chemistry-request@server.ccl.net Thu Apr 4 23:01:32 2002 Received: from ccl.net ([192.148.249.4]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g3541Wp08454 for ; Thu, 4 Apr 2002 23:01:32 -0500 Received: from arlen.ccl.net (arlen.ccl.net [192.148.249.10]) by ccl.net (8.11.6+Sun/8.11.6/OSC 2.0) with ESMTP id g3541U309054; Thu, 4 Apr 2002 23:01:32 -0500 (EST) Date: Thu, 4 Apr 2002 23:01:29 -0500 (EST) From: Jan Labanowski To: chemistry@ccl.net cc: Jan Labanowski Subject: CCL Coordinator needs your help -- I really do!!! Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Dear CCL Subscribers, I hope that by running this list I can occasionally ask you for your support. This time I am asking for two supports {:-)} Yes, I am shameless... 1) For CCL itself. Please help with your donations, purchase of CCL archives, and contributions to archives, There is also a new service: paying for job announcements. Check: http://www.ccl.net/chemistry/announcements/jobs (do not jump -- it is still free, but if you want to pay for it, I will provide you with the tools --- the quotations and invoices when needed). CCL needs your support... I am very busy lately, and the list needs constant attention, additions, and development. With the hostile Internet environment, it needs more attention to details than ever. This is done by my student coworkers -- you pay their salaries, so please help. The information how to support CCL is on the front page of CCL Web site: www.ccl.net 2) Please come to the meetings I am co-organizing on computational materials science. Check the Web site: http://asdn.net/emrs/fall2002 and the links to Spring E-MRS and Moscow Nano & Giga Conference. The meetings have already attracted a lot of interest and we have great speakers and topics lined up. Please come... Be a participant and submit a paper. If you are "in a position", consider encouraging your institution to sponsor, provide training, or have a booth at the E-MRS Fall 2002 meeting. This will be a highly visible event!!! I attach a short description of the E-MRS Fall 2002 Symposium A in Cracow. I am sorry if you saw it before, since I did recently a massive mailing about this meeting. Please distribute this information within your organization and with your colleagues if you think that they may be interested. THANK YOU!!! The European Materials Research Society Meetings are Europe's largest gatherings of material scientists and engineers. The Fall 2002 Meeting will be held in Cracow, Poland. Please consider being a part of this meeting by attending Symposium A: SOFTWARE DEVELOPMENT FOR PROCESS AND MATERIALS DESIGN. http://asdn.net/emrs/fall2002 Sept. 15-19 Cracow, Poland The E-MRS Meeting is traditionally very well attended with participants from Europe, Asia, and The Americas. We have already confirmed very good speakers and we are working on inviting even more high caliber presenters. Several sessions are planned spanning non-empirical and empirical approaches, at different scales for different materials needs. Check our Web site for details and start working on your abstract (the deadline is May 7, 2002 !!!). We will publish papers presented at the meeting with Elsevier. Details will be provided by the Meeting Organizers soon. While the emphasis of the symposium is on methodology development, papers which present excellent examples of applications are also invited. We plan to organize topical sessions and start each session from the overview of the current state-of-the-art. There are also other symposia at this meeting -- for topics check the Web site and links given there. Last but not least, the location of the meeting is very attractive. Cracow, a well known tourist attraction, is a beautiful historical city with charming atmosphere. While the meeting will be quite busy, there will be time to go to the Old Market Place or to see the Wawel Castle. This is a perfect time to be in Krakow with the summer crowds gone, and the beautiful colors of fall coming to the Planty -- the green belt which marks the location of old city walls. Please also consider supporting our meeting (details are available on the Web). This meeting would be a great opportunity to gain publicity and to present your products in this dynamically developing part of the world. If you know someone who may be interested in participating in this important meeting, or if you belong to the mailing list or newsgroup which can benefit from participating in this meeting, please forward this message. This will be a good and informative meeting. Thank you for your consideration. Jan Labanowski (jkl@ccl.net) for the Organizing Committee: Anatoli A. Korkin Jim Greer Stanislaw Kucharski anatoli.korkin@motorola.com jgreer@nmrc.ucc.ie sak@tc3.ich.us.edu.pl ------ Jan K. Labanowski | phone: 614-292-9279, FAX: 614-292-7168 Ohio Supercomputer Center | Internet: jkl@ccl.net 1224 Kinnear Rd, | http://www.ccl.net/chemistry.html Columbus, OH 43212-1163 | http://www.ccl.net/