From owner-chemistry@ccl.net Mon Oct 8 03:53:00 2012 From: "Simon Cross simon+/-moldiscovery.com" To: CCL Subject: CCL: PharmBench v1.0 Community Benchmarking Dataset Message-Id: <-47734-121008035054-25630-MnBHrtrGKrx3ioJ2o5B+3Q|,|server.ccl.net> X-Original-From: Simon Cross Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=ISO-8859-1; format=flowed Date: Mon, 08 Oct 2012 08:50:44 +0100 MIME-Version: 1.0 Sent to CCL by: Simon Cross [simon===moldiscovery.com] Hi Everyone - many thanks to those who pointed out that the link to downloading the complete dataset on the pharmbench.moldiscovery.com site was not working. This has now been fixed. Regards, Simon On 05/10/2012 11:54, Simon Cross simon _ moldiscovery.com wrote: > > Sent to CCL by: Simon Cross [simon],[moldiscovery.com] > Dear Colleagues, > > We would like to draw your attention to the publication of the > PharmBench 1.0 dataset for the evaluation of molecular alignment and > pharmacophore elucidation approaches. > > The dataset consists of 81 targets, containing 960 ligands in total. > The dataset is based on known pharmaceutically relevant > co-crystallised protein ligand complexes, which were filtered to leave > high-resolution structures containing drug-like small molecule > ligands, for which electron density is also available at the EDS > server. For each target the structures have been aligned using the > receptor backbone atoms, and then the ligands extracted and prepared > by fixing their atom types and adding hydrogen atoms. > > The datasets therefore contain target ligands, aligned in the > biologically relevant frame of reference, to act as a 'gold standard' > reference set. The two-dimensional structures of these ligands are > also provided. > > An ideal molecular alignment or pharmacophore elucidation method > should be able to read the two-dimensional input structures, and > output three-dimensional alignments of the ligands that are equivalent > to the gold standard alignments. In the coming weeks a web service > will be set up to enable user-provided alignment models to be > evaluated using the same approach as described in the publication > (dx.doi.org/10.1021/ci300154n). > > The dataset is available at pharmbench.moldiscovery.com. > > Kind regards, > > Simon > -- Dr. Simon Cross Snr Scientist & Product Manager Molecular Discovery Ltd +44 7980 572278 From owner-chemistry@ccl.net Mon Oct 8 05:20:00 2012 From: "Michel Petitjean petitjean.chiral||gmail.com" To: CCL Subject: CCL: PharmBench v1.0 Community Benchmarking Dataset Message-Id: <-47735-121008051855-24595-c1PLnQheUq39p9dpi+QvIQ[A]server.ccl.net> X-Original-From: Michel Petitjean Content-Type: text/plain; charset=ISO-8859-1 Date: Mon, 8 Oct 2012 11:18:47 +0200 MIME-Version: 1.0 Sent to CCL by: Michel Petitjean [petitjean.chiral+/-gmail.com] Dear Simon, I downloaded the first of your sets, P10828, which contains 3 molecules, and I looked to their common 3D motifs two by two with the help of CSR (http://petitjeanmichel.free.fr/itoweb.petitjean.freeware.html#CSR) I got the following for 10000 iterations (very fast), extracted from the results computed by CSR: MOL. 1 : 48 ATOMS ; 1n46_xtal_reference MOL. 2 : 41 ATOMS ; 2j4a_xtal_reference MOL. 3 : 29 ATOMS ; 2pin_xtal_reference 2 superposed on 1: 39 common atoms ; RMS = 0.329831 ; pairwise correspondence: 11 12 13 14 19 20 21 22 25 27 28 35 36 34 33 31 32 29 30 47 48 37 38 43 44 45 46 39 40 41 42 15 10 9 2 1 3 5 4 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 36 37 39 40 41 3 superposed on 1: 28 common atoms ; RMS = 0.343439 ; pairwise correspondence: 11 12 13 14 19 20 21 22 23 27 28 35 36 34 33 31 32 29 30 47 48 37 15 10 9 2 1 3 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 26 27 28 29 3 superpsoed on 2: 28 common atoms ; RMS = 0.407966 ; pairwise correspondence: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 28 32 33 34 36 37 39 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 26 27 28 29 The coordinates of the rotated and translated molecules are available, too, but I do not want to send a too long message (CSR is free, you can try it). Intersecting the lists, I would deduce that the pharmacophore contains 28 atoms. Is that the kind of calculus you offer with your web service ? Best regards, Michel. Michel Petitjean MTi, INSERM UMR-S 973, University Paris 7, CNRS SNC 9079 35 rue Helene Brion, 75205 Paris Cedex 13, France. Phone: +331 5727 8434; Fax: +331 5727 8372 E-mail: petitjean.chiral]=[gmail.com (preferred), michel.petitjean]=[univ-paris-diderot.fr http://petitjeanmichel.free.fr/itoweb.petitjean.freeware.html 2012/10/8 Simon Cross simon+/-moldiscovery.com : > > Sent to CCL by: Simon Cross [simon===moldiscovery.com] > Hi Everyone - many thanks to those who pointed out that the link to > downloading the complete dataset on the pharmbench.moldiscovery.com site was > not working. This has now been fixed. > > Regards, > > Simon > > On 05/10/2012 11:54, Simon Cross simon _ moldiscovery.com wrote: >> >> >> Sent to CCL by: Simon Cross [simon],[moldiscovery.com] >> Dear Colleagues, >> >> We would like to draw your attention to the publication of the PharmBench >> 1.0 dataset for the evaluation of molecular alignment and pharmacophore >> elucidation approaches. >> >> The dataset consists of 81 targets, containing 960 ligands in total. The >> dataset is based on known pharmaceutically relevant co-crystallised protein >> ligand complexes, which were filtered to leave high-resolution structures >> containing drug-like small molecule ligands, for which electron density is >> also available at the EDS server. For each target the structures have been >> aligned using the receptor backbone atoms, and then the ligands extracted >> and prepared by fixing their atom types and adding hydrogen atoms. >> >> The datasets therefore contain target ligands, aligned in the biologically >> relevant frame of reference, to act as a 'gold standard' reference set. The >> two-dimensional structures of these ligands are also provided. >> >> An ideal molecular alignment or pharmacophore elucidation method should be >> able to read the two-dimensional input structures, and output >> three-dimensional alignments of the ligands that are equivalent to the gold >> standard alignments. In the coming weeks a web service will be set up to >> enable user-provided alignment models to be evaluated using the same >> approach as described in the publication (dx.doi.org/10.1021/ci300154n). >> >> The dataset is available at pharmbench.moldiscovery.com. >> >> Kind regards, >> >> Simon >> > > > -- > Dr. Simon Cross > Snr Scientist & Product Manager > Molecular Discovery Ltd > +44 7980 572278 > From owner-chemistry@ccl.net Mon Oct 8 07:27:00 2012 From: "Simon Cross simon]-[moldiscovery.com" To: CCL Subject: CCL: PharmBench v1.0 Community Benchmarking Dataset Message-Id: <-47736-121008072533-25769-Sge8/9VnJh/tw35LDCb/OQ],[server.ccl.net> X-Original-From: Simon Cross Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=ISO-8859-1; format=flowed Date: Mon, 08 Oct 2012 12:25:25 +0100 MIME-Version: 1.0 Sent to CCL by: Simon Cross [simon|*|moldiscovery.com] Dear Michael, from what I understand from your documentation, CSR is doing this: CSR reads the cartesian coordinates of two molecules, then optimally rotates and translates the molecule 2 onto the molecule 1 to find the maximal common 3D motif. ie, the method is a way to align 1 molecule to a template. From your email below, it seems you are doing this in various combinations to get the common alignment. However, I guess you started from the 3D X-ray structures? This is only one part of the problem...we are trying to see how well we can reproduce the 3D X-ray alignments, but starting from the 2D structures. Our approach to do this is called FLAPpharm, and we just published this in part 1 of the paper I reported in the previous email (http://pubs.acs.org/doi/abs/10.1021/ci300153d), and it is based on GRID fields. The PharmBench dataset was something we put together to validate FLAPpharm. The idea of the web service is to score the resulting alignment models in a consistent manner, according to 3 metrics: 1. Can the method identify the correct X-ray conformation? This is simply an alignment of each predicted structure to the X-ray standard and rmsd calculation across the dataset. 2. Can the method identify the correct conf, and align them all to each other well? This is the AlignScore metric we implemented (initially described by Gareth Jones), where the alignment model is fitted to the 3D X-ray alignment, the rmsd calculated, and the number of successes reported. 3. Can the method produce a model with similar pharmacophoric interaction fields? Here we are using GRID to calculate the mean average Molecular Interaction Fields for the model, and calculating the field similarity to the X-ray alignment model. To summarise, the web service is not providing a pharmacophore elucidation service, but it should allow users to submit alignment models (hopefully starting from the 2D unbiased conformations), and score them in the same way in which we did in the paper to enable the comparison of methods. The metrics are not perfect, as we discuss in the paper, but I think it would be nice to have a 'standard' way of measuring pharmacophore elucidation and alignment tools. Of course, open to discussion! My best, Simon On 08/10/12 10:18, Michel Petitjean petitjean.chiral||gmail.com wrote: > Sent to CCL by: Michel Petitjean [petitjean.chiral+/-gmail.com] > Dear Simon, > > I downloaded the first of your sets, P10828, which contains 3 > molecules, and I looked to their common 3D motifs two by two with the > help of CSR > (http://petitjeanmichel.free.fr/itoweb.petitjean.freeware.html#CSR) > > I got the following for 10000 iterations (very fast), extracted from > the results computed by CSR: > > MOL. 1 : 48 ATOMS ; 1n46_xtal_reference > MOL. 2 : 41 ATOMS ; 2j4a_xtal_reference > MOL. 3 : 29 ATOMS ; 2pin_xtal_reference > > 2 superposed on 1: 39 common atoms ; RMS = 0.329831 ; pairwise correspondence: > 11 12 13 14 19 20 21 22 25 27 28 > 35 36 34 33 31 32 29 30 47 48 37 38 > 43 44 45 46 39 40 41 42 15 10 9 2 > 1 3 5 4 > 1 2 3 4 5 6 7 8 9 10 11 > 12 13 14 15 16 17 18 19 20 21 22 23 > 24 25 26 27 28 29 30 31 32 33 34 36 > 37 39 40 41 > > 3 superposed on 1: 28 common atoms ; RMS = 0.343439 ; pairwise correspondence: > 11 12 13 14 19 20 21 22 23 27 28 > 35 36 34 33 31 32 29 30 47 48 37 15 > 10 9 2 1 3 > 1 2 3 4 5 6 7 8 9 10 11 > 12 13 14 15 16 17 18 19 20 21 22 23 > 24 26 27 28 29 > > 3 superpsoed on 2: 28 common atoms ; RMS = 0.407966 ; pairwise correspondence: > 1 2 3 4 5 6 7 8 9 10 11 > 12 13 14 15 16 17 18 19 20 21 28 32 > 33 34 36 37 39 > 1 2 3 4 5 6 7 8 9 10 11 > 12 13 14 15 16 17 18 19 20 21 22 23 > 24 26 27 28 29 > > The coordinates of the rotated and translated molecules are available, > too, but I do not want to send a too long message > (CSR is free, you can try it). > > Intersecting the lists, I would deduce that the pharmacophore contains 28 atoms. > Is that the kind of calculus you offer with your web service ? > > Best regards, > > Michel. > > Michel Petitjean > MTi, INSERM UMR-S 973, University Paris 7, > CNRS SNC 9079 > 35 rue Helene Brion, 75205 Paris Cedex 13, France. > Phone: +331 5727 8434; Fax: +331 5727 8372 > E-mail: petitjean.chiral[-]gmail.com (preferred), > michel.petitjean[-]univ-paris-diderot.fr > http://petitjeanmichel.free.fr/itoweb.petitjean.freeware.html > > 2012/10/8 Simon Cross simon+/-moldiscovery.com : >> Sent to CCL by: Simon Cross [simon===moldiscovery.com] >> Hi Everyone - many thanks to those who pointed out that the link to >> downloading the complete dataset on the pharmbench.moldiscovery.com site was >> not working. This has now been fixed. >> >> Regards, >> >> Simon >> >> On 05/10/2012 11:54, Simon Cross simon _ moldiscovery.com wrote: >>> >>> Sent to CCL by: Simon Cross [simon],[moldiscovery.com] >>> Dear Colleagues, >>> >>> We would like to draw your attention to the publication of the PharmBench >>> 1.0 dataset for the evaluation of molecular alignment and pharmacophore >>> elucidation approaches. >>> >>> The dataset consists of 81 targets, containing 960 ligands in total. The >>> dataset is based on known pharmaceutically relevant co-crystallised protein >>> ligand complexes, which were filtered to leave high-resolution structures >>> containing drug-like small molecule ligands, for which electron density is >>> also available at the EDS server. For each target the structures have been >>> aligned using the receptor backbone atoms, and then the ligands extracted >>> and prepared by fixing their atom types and adding hydrogen atoms. >>> >>> The datasets therefore contain target ligands, aligned in the biologically >>> relevant frame of reference, to act as a 'gold standard' reference set. The >>> two-dimensional structures of these ligands are also provided. >>> >>> An ideal molecular alignment or pharmacophore elucidation method should be >>> able to read the two-dimensional input structures, and output >>> three-dimensional alignments of the ligands that are equivalent to the gold >>> standard alignments. In the coming weeks a web service will be set up to >>> enable user-provided alignment models to be evaluated using the same >>> approach as described in the publication (dx.doi.org/10.1021/ci300154n). >>> >>> The dataset is available at pharmbench.moldiscovery.com. >>> >>> Kind regards, >>> >>> Simon >>> >> >> -- >> Dr. Simon Cross >> Snr Scientist & Product Manager >> Molecular Discovery Ltd >> +44 7980 572278> > From owner-chemistry@ccl.net Mon Oct 8 08:24:00 2012 From: "Chris Swain swain]-[mac.com" To: CCL Subject: CCL: Viewing Docking results in Vortex using Astex Viewer Message-Id: <-47737-121008024204-21697-G1enabkfcY1aW/o2GzE/SQ**server.ccl.net> X-Original-From: Chris Swain Content-Transfer-Encoding: 8bit Content-type: text/plain; charset=windows-1252 Date: Mon, 08 Oct 2012 07:41:54 +0100 MIME-version: 1.0 Sent to CCL by: Chris Swain [swain- -mac.com] I just thought this may be of general interest. I’ve been working with Mike Hartshorn and the people at Dotmatics who have incorporated OpenAstexViewer a 3D molecule viewer into Vortex. http://macinchem.org/reviews/vortex2/vortex_docking.php Cheers, Chris Dr Chris Swain BA MA (Cantab) PhD CChem FRSC Macinchem Insanely Great Science swain{}mac.com http://www.macinchem.org From owner-chemistry@ccl.net Mon Oct 8 08:59:01 2012 From: "Sharan sara180681%gmail.com" To: CCL Subject: CCL:G: Polymer Construction Message-Id: <-47738-121008081900-15554-HeHgdGz0wQpta+b9wjWuIQ~~server.ccl.net> X-Original-From: Sharan Content-Type: multipart/alternative; boundary=f46d042dfc9550581304cb8b3953 Date: Mon, 8 Oct 2012 17:48:49 +0530 MIME-Version: 1.0 Sent to CCL by: Sharan [sara180681]-[gmail.com] --f46d042dfc9550581304cb8b3953 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: quoted-printable Hi Thomas Thanks for your valuable suggestion. Let me check with VESTA. My ultimate aim is to work on Vibration studies of Polymers. If you have any suggestion, please . With regards Saravanan On Sat, Oct 6, 2012 at 12:45 PM, Gkourmpis, Thomas Thomas.Gkourmpis a borealisgroup.com wrote: > Saravanan Hi**** > > ** ** > > I think we are talking about two completely different things here. If you > want to construct a polymer crystal then the method offered by GaussView = is > perfectly fine. If you want to construct an amorphous polymer (what you > call 3D structure) then you cannot use GaussView as this will create only > crystalline systems. In the latter case you need to construct the polymer > =93by hand=94 or using alternative software.**** > > ** ** > > Now regarding the problems you have with the PBC tutorials I would > recommend you don=92t waste your time and look for alternative software. > VESTA is a good possibility http://jp-minerals.org/vesta/en/ and you can > save the outputs as PDBs, but I am sure there are other packages (free or > otherwise available).**** > > ** ** > > I hope this helps**** > > Thomas**** > > ** ** > > *From:* owner-chemistry+thomas.gkourmpis=3D=3Dborealisgroup.com^^ccl.net[= mailto: > owner-chemistry+thomas.gkourmpis=3D=3Dborealisgroup.com^^ccl.net] *On Beh= alf > Of *Sharan sara180681]-[gmail.com > *Sent:* Friday, October 05, 2012 7:50 PM > *To:* Gkourmpis, Thomas > *Subject:* CCL:G: Polymer Construction**** > > ** ** > > Hi Gkourmpis Thomas**** > > **** > > At first i would like to thank, for you reply. **** > > **** > > In Gauss View using Periodic Boundary Conditions (PBC) one can construct > Polymer unit cell, after which super cell is constructed. This resembles > the whole polymer, is what my understanding. I don't think so the constra= in > of, number of repeating units comes into picture.**** > > **** > > **** > > But my problem is the steps given for building the unit cell is not clear > to me. There are 20 steps, using these steps i want to construct a polyme= r > of my interest.**** > > **** > > When you say "construct by hand are you referring to PBC or just regular > 3D structure of a monomer?**** > > **** > > With regards**** > > Saravanan**** > > On Thu, Oct 4, 2012 at 4:02 PM, Gkourmpis, Thomas Thomas.Gkourmpis## > borealisgroup.com wrote:**** > > Hi**** > > **** > > Regarding the polymer in Gaussian the question is how many repeating unit= s > you need. If you need only a few units and each is small you can always > construct it by hand in GaussView. If on the other hand you want a =93pro= per=94 > chain with many units then you=92ll be better off if you use any of the > builders used in Molecular Dynamics packages and construct your system > there. Then you can convert it to any file that Gaussian can read and use > it as you see fit.**** > > **** > > Have I answered your question or I misunderstood something?**** > > **** > > Thomas**** > > **** > > *From:* owner-chemistry+thomas.gkourmpis=3D=3Dborealisgroup.com-x-ccl.net= [mailto: > owner-chemistry+thomas.gkourmpis=3D=3Dborealisgroup.com-x-ccl.net] *On Be= half > Of *Sharan sara180681%%gmail.com > *Sent:* Thursday, October 04, 2012 10:59 AM > *To:* Gkourmpis, Thomas > *Subject:* CCL:G: Polymer Construction**** > > **** > > Hi **** > > **** > > Can somebody help me in constructing Polymeric unit in Gaussian for > optimization. I am not clear on constructing the Translational vector for > polymeric unit. I went through the example given in Gaussian 03, I am not > clear on few fixed values supplied in the examples in constructing the > polymeric unit. **** > > **** > > Some helping hand will be greateful.. thanks in advance.**** > > **** > > With best regards**** > > **** > > **************************************************** > Dr. I. SARAVANAN, M.Sc., M.Phil., PGDCA., PGDHRM., Ph.D., > Consultant, ***** > > This Email and any files transmitted with it are confidential and intende= d > solely for the use of the individual or the entity to whom it is addresse= d. > If you have received this Email by error, please notify the sender and > delete the material from any storage device. Borealis extends no warranti= es > and makes no representations as to the accuracy or completeness of the > information provided. It is the customer's responsibility to inspect and > test our products and technical advice in order to satisfy itself as to t= he > suitability of the products and technical advice for the customer's > particular purpose. **** > > > > > --f46d042dfc9550581304cb8b3953 Content-Type: text/html; charset=windows-1252 Content-Transfer-Encoding: quoted-printable

Hi Thomas
=A0
Thanks for your valuable suggest= ion. Let me check with VESTA. My ultimate aim is to work on Vibration studi= es of Polymers. If you have any suggestion, please .
=A0
With regards
Saravanan
=A0
=A0
On Sat, Oct 6, 2012 at 12:45 PM, Gkourmpis, Thomas Thomas= .Gkourmpis a borealisgroup.com <owner-chemistry ~~ ccl.net> wrote:
Saravanan Hi

=A0

I think we are talki= ng about two completely different things here. If you want to construct a p= olymer crystal then the method offered by GaussView is perfectly fine. If y= ou want to construct an amorphous polymer (what you call 3D structure) then= you cannot use GaussView as this will create only crystalline systems. In = the latter case you need to construct the polymer =93by hand=94 or using al= ternative software.

=A0

Now regardin= g the problems you have with the PBC tutorials I would recommend you don=92= t waste your time and look for alternative software. VESTA is a good possib= ility http:/= /jp-minerals.org/vesta/en/ and you can save the outputs as PDBs, but I = am sure there are other packages (free or otherwise available).

=A0

I hope this = helps

Thomas=

= =A0

From: owner-chemistry+thomas.gkourmpis=3D=3Dborealisgroup.com^^ccl.net [mailto:owner-chemistry+thomas.gkour= mpis=3D=3Dboreal= isgroup.com^^ccl.net] = On Behalf Of Sharan sara180681]-[gmail.com
Sent: Friday, October 05, 2012 7:50 PM
To: Gkourmpis, Thom= as
Subject: CCL:G: Polymer Construction

<= /div>

=A0

Hi Gkourmpis Thomas

=A0<= u>

At first i would like to= thank, for you reply.

= =A0

In Gauss View using Periodic Boundary Con= ditions (PBC) one can construct Polymer unit cell, after which super cell i= s constructed. This resembles the whole polymer, is what my understanding. = I don't think so the constrain of, number of repeating units comes into= picture.

=A0

=A0

But m= y problem is the steps given for building the unit cell is not clear to me.= There are 20 steps, using these steps=A0i want to construct a polymer of m= y interest.

=A0

When you say "construct by hand are you referring to PB= C or just regular 3D structure of a monomer?

=A0

With regards<= u>

Sa= ravanan

On Thu, Oct 4, 2= 012 at 4:02 PM, Gkourmpis, Thomas Thomas.Gkourmpis##borealisgroup.com <owner-chemistry=CCl.net>= wrote:

Hi

=A0

Regard= ing the polymer in Gaussian the question is how many repeating units you ne= ed. If you need only a few units and each is small you can always construct= it by hand in GaussView. If on the other hand you want a =93proper=94 chai= n with many units then you=92ll be better off if you use any of the builder= s used in Molecular Dynamics packages and construct your system there. Then= you can convert it to any file that Gaussian can read and use it as you se= e fit.

=A0

Have I answered your question or I misunder= stood something?

=A0

Thomas=

=A0=

From: owner-chemistry+thomas.gkourmpis=3D=3Dborealisgroup.com-x-ccl.net [mailto:owner-chemistry+thomas.gkourmpis=3D=3Dborealisgroup.com-x-ccl.net] On Behalf Of Sharan sara180681%%gmail.com
Sent: Thursday, October 04, 2012 10:59 AM
To: Gkourmpis, T= homas
Subject: CCL:G: Polymer Construction

=A0

Hi

=A0

Can somebody help me in constructing Pol= ymeric unit in Gaussian for optimization. I am not clear on constructing th= e Translational vector for polymeric unit. I=A0went through the example giv= en in Gaussian 03, I am not clear on few fixed values supplied in the examp= les in constructing the polymeric unit.

=A0

Some helping hand will be greateful.. thanks in advance.<= /u>

=A0

=

With best regards

=A0=

******************= *******************************
Dr. I. SARAVANAN, M.= Sc., M.Phil., PGDCA., PGDHRM., Ph.D.,
Consultant,

=

This Email and any files transmitted with it a= re confidential and intended solely for the use of the individual or the en= tity to whom it is addressed. If you have received this Email by error, ple= ase notify the sender and delete the material from any storage device. Bore= alis extends no warranties and makes no representations as to the accuracy = or completeness of the information provided. It is the customer's respo= nsibility to inspect and test our products and technical advice in order to= satisfy itself as to the suitability of the products and technical advice = for the customer's particular purpose.




--f46d042dfc9550581304cb8b3953-- From owner-chemistry@ccl.net Mon Oct 8 09:34:00 2012 From: "Michel Petitjean petitjean.chiral#,#gmail.com" To: CCL Subject: CCL: PharmBench v1.0 Community Benchmarking Dataset Message-Id: <-47739-121008092142-5600-Usu0dEtqcuEEaU8x86m3Dw^server.ccl.net> X-Original-From: Michel Petitjean Content-Type: text/plain; charset=ISO-8859-1 Date: Mon, 8 Oct 2012 15:21:37 +0200 MIME-Version: 1.0 Sent to CCL by: Michel Petitjean [petitjean.chiral*_*gmail.com] Dear Simon, You are right, CSR is just a 3D alignment tool. According to your last sentence, it would be indeed great to "measure alignment tools", i.e. define criteria to measure the quality of 3D alignments. For that, we need to decide when a 3D alignment programme is successfull and when it fails. But "who" can decide that ? The chemist ? An other alignment algorithm ? It is a part of the problem, and compared to 2D alignments, it is not so simple to answer this question. Each of the many 3D existing alignment tools has its own criteria, and most time it involves various parameters to be optimized in function of the input data, so that there is not one alignment to be computed for one pair of 3D molecules, but there are as so many 3D alignments that there are values of these parameters (e.g., a cut-off value for the RMS, etc.). Remark: CSR has no such parameter. The discussion is indeed opened. Thanks. Michel. 2012/10/8 Simon Cross simon]-[moldiscovery.com : > > Sent to CCL by: Simon Cross [simon|*|moldiscovery.com] > Dear Michael, from what I understand from your documentation, CSR is doing > this: > > CSR reads the cartesian coordinates of two molecules, then optimally rotates > and > translates the molecule 2 onto the molecule 1 to find the maximal common 3D > motif. > > ie, the method is a way to align 1 molecule to a template. From your email > below, it seems you are doing this in various combinations to get the common > alignment. However, I guess you started from the 3D X-ray structures? This > is only one part of the problem...we are trying to see how well we can > reproduce the 3D X-ray alignments, but starting from the 2D structures. > > Our approach to do this is called FLAPpharm, and we just published this in > part 1 of the paper I reported in the previous email > (http://pubs.acs.org/doi/abs/10.1021/ci300153d), and it is based on GRID > fields. > > The PharmBench dataset was something we put together to validate FLAPpharm. > The idea of the web service is to score the resulting alignment models in a > consistent manner, according to 3 metrics: > > 1. Can the method identify the correct X-ray conformation? This is simply an > alignment of each predicted structure to the X-ray standard and rmsd > calculation across the dataset. > > 2. Can the method identify the correct conf, and align them all to each > other well? This is the AlignScore metric we implemented (initially > described by Gareth Jones), where the alignment model is fitted to the 3D > X-ray alignment, the rmsd calculated, and the number of successes reported. > > 3. Can the method produce a model with similar pharmacophoric interaction > fields? Here we are using GRID to calculate the mean average Molecular > Interaction Fields for the model, and calculating the field similarity to > the X-ray alignment model. > > To summarise, the web service is not providing a pharmacophore elucidation > service, but it should allow users to submit alignment models (hopefully > starting from the 2D unbiased conformations), and score them in the same way > in which we did in the paper to enable the comparison of methods. > > The metrics are not perfect, as we discuss in the paper, but I think it > would be nice to have a 'standard' way of measuring pharmacophore > elucidation and alignment tools. > > Of course, open to discussion! > > My best, > > Simon > From owner-chemistry@ccl.net Mon Oct 8 10:40:00 2012 From: "Simon Cross simon(a)moldiscovery.com" To: CCL Subject: CCL: PharmBench v1.0 Community Benchmarking Dataset Message-Id: <-47740-121008103848-28266-5X2M5ZuBT3nrvCMp8dXGxQ{}server.ccl.net> X-Original-From: Simon Cross Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=ISO-8859-1; format=flowed Date: Mon, 08 Oct 2012 15:38:40 +0100 MIME-Version: 1.0 Sent to CCL by: Simon Cross [simon###moldiscovery.com] Dear Michel, The "who can decide that" question you pose - for me it is the experimental data. The point of the PharmBench dataset is to provide "correct" alignments of the 3D ligands so we can measure the methods objectively using the metrics I described. "correct" in this case is coming from protein backbone alignments for one family, and subsequently extracting the ligand coordinates and preparing them (atom typing, tautomerisation etc). So the alignments are coming from experimental data. Of course there is some error coming from the protein alignment, the resolution of the data etc, but in general we think it's a good place to start. Cheers, Simon On 08/10/12 14:21, Michel Petitjean petitjean.chiral#,#gmail.com wrote: > Sent to CCL by: Michel Petitjean [petitjean.chiral*_*gmail.com] > Dear Simon, > You are right, CSR is just a 3D alignment tool. > According to your last sentence, it would be indeed great to "measure > alignment tools", > i.e. define criteria to measure the quality of 3D alignments. > For that, we need to decide when a 3D alignment programme is > successfull and when it fails. > But "who" can decide that ? The chemist ? An other alignment algorithm ? > It is a part of the problem, and compared to 2D alignments, it is not > so simple to answer this question. > Each of the many 3D existing alignment tools has its own criteria, > and most time it involves various parameters to be optimized in > function of the input data, > so that there is not one alignment to be computed for one pair of 3D molecules, > but there are as so many 3D alignments that there are values of these parameters > (e.g., a cut-off value for the RMS, etc.). > Remark: CSR has no such parameter. > The discussion is indeed opened. > Thanks. > Michel. > > 2012/10/8 Simon Cross simon]-[moldiscovery.com : >> Sent to CCL by: Simon Cross [simon|*|moldiscovery.com] >> Dear Michael, from what I understand from your documentation, CSR is doing >> this: >> >> CSR reads the cartesian coordinates of two molecules, then optimally rotates >> and >> translates the molecule 2 onto the molecule 1 to find the maximal common 3D >> motif. >> >> ie, the method is a way to align 1 molecule to a template. From your email >> below, it seems you are doing this in various combinations to get the common >> alignment. However, I guess you started from the 3D X-ray structures? This >> is only one part of the problem...we are trying to see how well we can >> reproduce the 3D X-ray alignments, but starting from the 2D structures. >> >> Our approach to do this is called FLAPpharm, and we just published this in >> part 1 of the paper I reported in the previous email >> (http://pubs.acs.org/doi/abs/10.1021/ci300153d), and it is based on GRID >> fields. >> >> The PharmBench dataset was something we put together to validate FLAPpharm. >> The idea of the web service is to score the resulting alignment models in a >> consistent manner, according to 3 metrics: >> >> 1. Can the method identify the correct X-ray conformation? This is simply an >> alignment of each predicted structure to the X-ray standard and rmsd >> calculation across the dataset. >> >> 2. Can the method identify the correct conf, and align them all to each >> other well? This is the AlignScore metric we implemented (initially >> described by Gareth Jones), where the alignment model is fitted to the 3D >> X-ray alignment, the rmsd calculated, and the number of successes reported. >> >> 3. Can the method produce a model with similar pharmacophoric interaction >> fields? Here we are using GRID to calculate the mean average Molecular >> Interaction Fields for the model, and calculating the field similarity to >> the X-ray alignment model. >> >> To summarise, the web service is not providing a pharmacophore elucidation >> service, but it should allow users to submit alignment models (hopefully >> starting from the 2D unbiased conformations), and score them in the same way >> in which we did in the paper to enable the comparison of methods. >> >> The metrics are not perfect, as we discuss in the paper, but I think it >> would be nice to have a 'standard' way of measuring pharmacophore >> elucidation and alignment tools. >> >> Of course, open to discussion! >> >> My best, >> >> Simon> >