#235 - Co-sponsored Sessions

ACS National Meeting
Spring, 2008
New Orleans, LA

SUNDAY MORNING

COMP - Drug Discovery: Docking and Scoring
Ernest N. Morial Convention Center 348
I. Visiers, Organizer; A. Tebben, Presiding
8:00 12 Abandoning the rigid receptor approximation: Side-chain flexibility in GOLD
Peter Carlqvist, carlqvist@ccdc.cam.ac.uk, Cambridge Crystallographic Data Centre, 12 Union Road, CB2 1EZ Cambrisge, United Kingdom and John W Liebeschuetz, john@ccdc.cam.ac.uk, Applications Group, Cambridge Crystallographic Data Centre, 12 Union Rd, Cambridge, CB2 1EZ, United Kingdom

In the rapid and ongoing development of ligand docking methods a recent goal has been to abandon the use of a single ‘low'-energy structure of the receptor. It has long been clear that Fischer's famous lock and key concept, although attractive, is an oversimplification of reality - Especially in the field of drug discovery where the goal is often to disrupt an enzyme's function: as the key changes so may the lock!

The use of fully flexible ligands is considered standard for all ‘state of the art' docking programs, and most programs are fast and efficient in finding the correct poses, even if the search for the perfect energy function continues. Today most programs also include a way to treat receptor rearrangement upon ligand binding (induced fit). Methods range from ‘simple' solutions such as soft potentials to more sophisticated methods relying on homology modelling packages. The problem is complex due to the potentially large conformational space coupled with the requirement for rapid computations.

In this study we have used the recently implemented flexible side-chains option in the docking program GOLD to probe induced fit in important drug targets. Cross-docking has been performed on publicly available protein structures together with virtual screening using the challenging DUD-set to demonstrate the potential usefulness and pitfalls of docking with flexible receptors in modern-day drug discovery.

8:25 13 Accurate docking and scoring of fragment molecules for lead discovery and optimization
Kathryn Armstrong and B. Woody Sherman, Woody.Sherman@schrodinger.com. Schrödinger, Inc, 120 West 45th Street, New York, NY 10036-4041

In previous work we described a method to generate structure-based pharmacophore hypotheses derived from fragment docking. In this work we focus on methods to accurately dock and score fragments. While fragments generally bind weakly, we find that accounting for receptor flexibility is still important and show that our Induced-fit Docking methodology can accurately predict binding modes of fragments and the conformational reorganization of the protein. Next, we look at database virtual screening of fragments and show that it is possible to obtain substantial enrichment of actives, although not on par with virtual screening studies of drug-like compounds. We discuss the reasons for this and why false positives in computational fragment screening are more common and less problematic for drug discovery. Finally, we apply methods such as fragment joining and fragment growing to generate drug-like compounds from initial fragment hits.

8:50 14 Advances in induced-fit docking with applications toward predicting binding energies of diverse molecules
Ian Alberts, alberts@schrodinger.com1, Robert B. Murphy, murphy@schrodinger.com2, B. Woody Sherman, Woody.Sherman@schrodinger.com3, Ramy Farid3, and Richard A. Friesner4. (1) Schrodinger, New York, MA 10036, (2) Schrödinger, 120 W. 45th Street, New York, NY 10036, (3) Schrödinger, Inc, 120 West 45th Street, New York, NY 10036, (4) Department of Chemistry, Columbia University, 116th and Broadway, New York, NY 10027

The ability to accurately predict the structure of a flexible ligand-receptor complex is a key challenge in computational drug discovery. Obtaining an accurate structure can provide insights into important interactions that drive ligand binding and is necessary to predict binding energies. In this work we present recent advances in our induced-fit docking methodology. Significant improvements to accuracy and speed have been made by incorporating an adaptive softening potential that allows key receptor residues detected by the algorithm to be fully flexible while more rigid parts are treated with less flexibility. We present a much more extensive set of cross-docking cases covering a broad range of targets and ligands, and show that results from the new method are substantially improved over our previous induced-fit docking results. Finally, we couple the induced-fit structures with a new version of Glide XP to obtain significant correlations between predicted and experimental binding free energies.

9:15 15 Improved water handling in structure-based molecular docking
Angela Rumpl, Carsten Detering, detering@biosolveit.de, and Holger Claußen, Holger.Claussen@biosolveit.de. BioSolveIT GmbH, An der Ziegelei 75, 53757 Sankt Augustin, Germany

The proper handling of water in active sites is still a largely unsolved problem in structure-based drug design. We propose a novel strategy that allows for water molecules to be taken into consideration during placement generation in docking. The water can be either placed manually and fixed in orientation or kept rotatable, so that according to the ligand poses, it can best possibly form interactions with the protein and the small molecule. Also, the water can be made permanent or displaceable, so that the ligand placements and scores determine whether the water stays or is removed. This treatment is important for modelling buried or conserved water that is replaced by stronger binding ligands. Finally, aside of the manual placement, FlexX can suggest potential water positions which are then subjected to either of the above ways of handling such water. We present this novel technology together with application examples, illustrating the usefulness of the procedure.

9:40   Intermission
9:55 16 Predicting absolute binding free energies with physics-based methods
David L. Mobley, dmobley@maxwell.compbio.ucsf.edu, Pharmaceutical Chemistry, University of California, San Francisco, 600 16th Street, Box 2240, San Francisco, CA 95616 and Ken A. Dill, dill@maxwell.compbio.ucsf.edu, Department of Pharmaceutical Chemistry, University of California at San Francisco, San Francisco, CA 94143

Accurate and reliable predictions of binding free energies would be a tremendous aid to drug design, but this has proven extremely challenging. We discuss recent work applying rigorous alchemical free energy calculations to compute absolute binding free energies in a predictive context in several different model binding sites. We highlight lessons learned, including the relative contributions of multiple ligand orientations, protein conformational changes, and protein flexibility. We also present highlights from work in progress on applying these techniques to realistic binding sites.

10:20 17 Protein loop flexibility around ligand binding sites: Implications for drug design
Carolyn A. Weigelt, carolyn.weigelt@bms.com, Karen A. Rossi, karen.rossi@bms.com, Akbar Nayeem, akbar.nayeem@bms.com, and Stanley R. Krystek Jr., stanley.krystek@bms.com. Computer-Assisted Drug Design, Bristol-Myers Squibb Company, P.O. Box 5400, Princeton, NJ 08543-5400

A common procedure in deriving binding models for ligand-protein interactions is to treat the protein rigidly while allowing ligand flexibility. While this often gives reasonable binding models, there are numerous instances where the size and shape of the protein binding pocket changes quite dramatically upon ligand binding. Such structural changes are, as expected, most significant in the ″structurally variable regions″ of a protein, most commonly the loops. This study shows a variety of therapeutically interesting proteins that exhibit significant loop conformational changes upon binding different ligands. The drawbacks of generating binding models using ″incorrect″ loop geometries is discussed, underscoring the need to consider an ensemble of loop conformations rather than a single structure for ligand docking.

10:45 18 Role of quantum mechanical energies in binding sites of metalloproteins
Art E. Cho, artcho@korea.ac.kr, Department of Bioinformatics and Biotechnology, Korea University, Jochiwon-Eup, Yeongi-Gun, Chungnam, South Korea and David Rinaldo, Schrodinger, Inc, 120 W. 45th St., New York, NY 10036

As commonly acknowledged, binding between protein and other molecules is a fundamental phenomenon in biological processes. A protein binding could involve a number of different interactions including Coulombic, van der Waals, and hydrogen bond, all of which have their origin in electric charge. In order to model these processes on a computer, one needs to use the right blend of theories. Force field based molecular mechanics has primarily been used to simulate protein binding. However, in some cases, especially those involving highly polarized binding sites, fixed charge model molecular mechanics fails to describe binding modes of two molecules accurately. Quantum mechanical / molecular mechanical (QM/MM) method has gained popularity in last few years for description of such systems since one can use quantum mechanical level theories for only part of the system in question to obtain better description and yet maintain reasonable computational cost. In particular, when electron transfer is suspected within the binding site of a protein, quantum mechanical theory is needed to fully understand the binding process. We use QM/MM method to study systems in which electron transfer occurs, such as metalloproteins. We also devise a docking protocol which covers extended regions with quantum mechanics and apply it to various systems. The results show that the employment of QM/MM methods greatly improves prediction of binding modes in a few classes of proteins, including metalloproteins.

11:10 19 Structure-based lead optimization of small molecule β-secretase (BACE1) inhibitors
Kristi Yi Fan1, Jonanthan Bard2, Rajiv Chopra3, Derek Cole3, Jim Erdei3, William F. Fobare, fobarew@wyeth.com3, Iwan Gunawan3, Yun Hu, Christine Humblet3, Eric S. Manas3, Andrea Olland4, Nowak Pawel3, Dominick A. Quagliato, quaglid@wyeth.com3, Peter Reinhart, reinhap@wyeth.com2, William R. Solvibile3, William S Somers5, Jim Turner2, Eric Wagner2, Yinfa Yan, yany1@wyeth.com3, Ping Zhou, zhoup@wyeth.com3, Albert J. Robichaud, robicha@wyeth.com1, and Michael S. Malamas3. (1) Chemical Sciences, Wyeth Research, CN8000, Princeton, NJ 08543, (2) Neuroscience, Wyeth Research, CN 8000, Princeton, NJ 08543, (3) Chemical and Screening Sciences, Wyeth Research, MA, (4) Chemical & Screening Sciences, Wyeth Research, 87 Cambridge Park Dr, Cambridge, MA 02140, (5) Chemical and Screening Sciences, Wyeth Research, 200 Cambridge Park Drive, Cambridge, MA 02140

Alzheimer's Disease (AD) is a progressive neurodegenerative disease and the leading cause of dementia. It is believed that the deposition of β-amyloid peptide (Aβ) resulting from the cleavage of Amyloid Precursor Protein (APP) may cause the development of of AD. β-secretase (BACE1) is a membrane-tethered aspartyl protease that cleaves the β-amyloid precursor protein (APP) and generates the N-terminus of Aβ, and therefore the reduction of brain Aβ levels through the inhibition of BACE1 is being pursued as an attractive approach for AD therapy. Despite extensive research over the past decade, it is only recently that high-affinity small molecule BACE1 inhibitors have emerged. We present here the optimization of a weak HTS lead WY-24454 (IC50 = 40 uM) to highly potent and selective BACE1 inhibitors through an iterative structure-based approach with the aid of the X-ray crystallography and molecular modeling. A variety of computational methods were applied in this endeavor to assist structure-based lead optimization effort, ranging from GRID analysis, quantum mechanical studies, QSAR modeling, and molecular docking with improved scoring function. The approach enabled us to identify key interactions in the enzyme site, to understand the role of the FLAP region (with regard to selectivity over BACE2), and to assess the physical parameters for improving cell-based activity and brain penetration. This approach enabled us to rapidly explore the rather large ligand-binding pocket of BACE1 and produce structurally diverse, and highly potent (IC50 ~10 nM) BACE1 inhibitors. These potent and selective BACE1 inhibitors will potentially lead to the identification of disease-modifying AD therapeutics.

MONDAY MORNING

COMP - Drug Discovery: Finding Hits
Ernest N. Morial Convention Center 348
I. Visiers, Organizer; J. Woolfrey, Presiding
8:00 54 Identification of Pyk2 FERM ligands by combining protein similarity assessment, mutagenesis study, pharmacophore prediction, and in silico screening
Lei Wang, lwang@tripos.com1, Nathalie Meurice, NMeurice@tgen.org2, Joseph Loftus2, Yuan-Ping Pang, pang@mayo.edu3, Bob Clark, bclark@tripos.com4, and Christopher A. Lipinski5. (1) Tripos Informatics Research Center, 1699 S. Hanley Rd., St. Louis, MO 63144, (2) Pharmaceutical Genomics Division, Translational Genomics Research Institute (TGen), 13208 E. Shea Blvd., Suite 110, Scottsdale, AZ 85259, (3) Mayo Clinic Cancer Center, Mayo Foundation for Medical Education and Research, 200 First Street SW, Rochester, MN 55905, (4) Tripos, Inc, 1699 S. Hanley Rd., St. Louis, MO 63144, (5) Mayo Clinic Scottsdale, 13400 East Shea Boulevard, Scottsdale, AZ 85259

The strong tendency of malignant glioma cells to invade locally into surrounding normal brain precludes effective surgical resection and reduces the efficacy of radiotherapy, Pyk2 (proline-rich tyrosine kinase 2) contributes to in vitro glioma migration and disease progression in vivo. The N-terminal FERM domain is functionally critical for Pyk2-mediated effector signaling. A three-dimensional model of Pyk2 was generated (PDB: 2FO6) and compared with available bound structures of related FERM domains. Important Pyk2 FERM residues were identified by similarity and confirmed by mutagenesis study. A protein pharmacophore model was subsequently created and used to search the LeadQuest small molecule database. Molecules identified in the screen were refined by docking. Top compounds (n=67) were screened by competition ELISA using an active site-specific antibody targeting the Pyk2 FERM. This methodology yielded 9 confirmed actives and validated this combined approach for identifying Pyk2 FERM ligands.

8:25 55 Improving enrichment rates: A practical solution to an impractical problem
Noel M. O'Boyle, oboyle@ccdc.cam.ac.uk, Cambridge Crystallographic Data Centre, 12 Union rd, Cambridge, CB2 1EZ, United Kingdom and Robin Taylor, robin.t@virgin.net, 54 Sherfield Avenue, Rickmansworth WD3 1NL, UK, United Kingdom

The weights of the terms in many of the empirical scoring functions used for docking were trained by regression against binding affinity (Ki) values for a set of protein-ligand complexes. Although it would indeed be very useful to be able to calculate accurate binding energies, the practical problem in protein-ligand docking is simply to distinguish active molecules from inactives. This suggests that information on inactive poses, or negative data, should be incorporated into the training procedure.

We have recently introduced a scaling function into the ChemScore scoring function used by GOLD, which is based on the burial depth of an interaction in the binding site. By training using negative data, we show that the discrimination between active and inactive molecules is greatly increased

8:50 56 Annotated DB of Chemically Feasible Scaffolds: Key Point for an Efficient Scaffold Hopping
Julen Oyarzabal, joyarzabal@cnio.es1, Trevor Howe2, Jose Ignacio Andres3, Jesus Alcazar3, and Rosa Maria Alvarez4. (1) Molecular Informatics, Johnson & Johnson Pharmeaceutical R&D. Current address: Spanish National Cancer Research Centre (CNIO), Jarama 75, Toledo, 45007, Spain, (2) Molecular Informatics, Johnson & Johnson Pharmaceutical Research & Development, Turnhoutseweg 30, 2340, Beerse, Belgium, (3) Department of Medicinal Chemistry, Johnson & Johnson Pharmaceutical R&D, Jarama, 75, Toledo, 45007, Spain, (4) Department of Medicinal Chemistry, Johnson & Johnson Pharmaceutical R&D. Current address: Spanish National Cancer Research Centre (CNIO), Jarama, 75, Toledo, 45007, Spain

Our main goal in drug discovery is reaching clinical phases with a good candidate in terms of PK/PD and with a solid IP position. Along this process there are too many barriers to overcome, from chemical feasibility, ADME related issues… to IP status. Therefore, looking for alternative chemotypes to main chemical series, its bioisosteric replacement, is a highly recommended approach if we want to achieve our goal.

Ideas for addressing the scaffold hopping in the most efficient manner are presented. Key starting point is building up an annotated DB with hundreds of thousands of unique, and chemically feasible, scaffolds. Details describing the process to build and use this DB will be discussed.

In addition, a case study where this approach was applied, providing a real added value to a drug discovery project, is presented

9:15 57 Fragment based de novo design using an existing fragment based docking program, eHiTS
Darryl Reid1, Zsolt Zsoldos1, and A Peter Johnson, a.p.johnson@chemistry.leeds.ac.uk2. (1) SimBioSys Inc, 135 Queen's Plate Dr, Suite 520, Toronto, ON M9W 6V1, Canada, (2) School of Chemistry, University of Leeds, Leeds, LS2 9JT, United Kingdom

Fragment based drug discovery has become a hot topic in recent years. The eHiTS docking program has been using a fragment based approach to docking since its inception. The algorithm divides input molecules into fragments and docks each fragment independently of the others before reconnecting to re-form the input molecule. Applying this methodology to fragment based de novo ligand design is a natural extension resulting in an efficient new tool. BACE-1 is a well studied flexible enzyme with implications in the treatment of Alzheimer's disease and over 30 public crystal structures. Using a set of co-crystallized ligands, we fragment the ligands and dock the fragments. By comparing the fragment poses to the original ligand pose, we can validate the ability of this protocol to reproduce known binders from fragments. It will be demonstrated how fragment docking results from various known ligands can be recombined and linked with additional common fragments to form some novel potential BACE-1 inhibitors.

9:40   Intermission
9:55 58 Identification of a potent novel non-steroidal progesterone receptor modulator from a virtual screen
Ray J. Unwalla1, Andrew Fensome1, Michael A. Marella1, Jason Cross1, Edward G. Melenski1, James Wilhelm2, Andrea Olland3, Scott Wolfrom2, Hassan Elokdah2, Jay Wrobel1, Richard C. Winneker4, Matthew R. Yudt4, Sunil Nagpal5, and Jeff Cohen4. (1) Chemical Sciences, Wyeth Research, 500 Arcola Rd., Collegeville, PA 19426, (2) Chemical and Screening Sciences, Wyeth Research, Cambridge, MA 02140, (3) Chemical & Screening Sciences, Wyeth Research, 87 Cambridge Park Dr, Cambridge, MA 02140, (4) Musculoskeletal Therapies, Wyeth Research, 500 Arcola Road, Collegeville, PA 19426, (5) Women's Health and Musculoskeletal Biology, Wyeth Research, 500 Arcola Road, Collegeville, PA 19426

Progesterone plays an important role in the regulation of female reproductive functions. Synthetic steroidal progestins are widely used as oral contraceptives and for treating a variety of other endocrinological diseases and disorders, however, concerns of side effects due to cross reactivity with other steroid hormone receptors such as the glucocorticoid receptor (GR) and androgen receptor (AR), have prompted us to search for novel non-steroidal progesterone modulators with improved receptor and tissue selectivity. A virtual screen of the Available Chemical Database was performed using in-house docking program i.e PharmDOCK on the x-ray structure of the human progesterone receptor(PR) and after applying physical properties and undesirable chemical functionality filters, 103 compounds were submitted for testing in the T47D alkaline phosphatase assay. This approach proved to be successful with 10 of the 103 compounds showing PR antagonist activity with Ic50 values ranging from 5 nM to 500 nM (~10% hit rate). In this presentation we will discuss the identification and characterization of one of the hits. We will also discuss our approach to build a nuclear hormone focus-screening library using ligand-based virtual screening of published x-ray structures.

10:20 59 Virtual screening discovered catechol-containing compounds as STAT3 SH2 domain inhibitors
Yongbo Hu1, Wenshan Hao2, Xinyi Huang3, Chao-Pei Betty Chang2, Jay Gibbons2, Jun Xu2, and Christine Humblet3. (1) Department of Structural Biology & Computational Chemistry, Wyeth Research, Pearl River, NY 10965, (2) Department of Oncology, Wyeth Research, Pearl River, NY 10965, (3) Chemical and Screening Sciences, Wyeth Research, Collegeville, PA 19426

Excessive activation of STAT3 (Signal Transducers and Activators of Transcription 3) has been correlated with a wide spectrum of cancers. Activation of STAT3 requires homodimerization between a pTyr (phosphotyrosine)-containing peptide of one monomer and the SH2 domain of another. To find leads from which to develop STAT3 SH2 inhibitors as anti-cancer agents, a proprietary Wyeth library of 112,386 lead-like compounds was screened. Compounds were docked to the STAT3 SH2 domain, using Glide SP. The top 1,000 virtual screening hits were tested in biochemical assays and 74 competitive STAT3 SH2 inhibitors were found. Among the active compounds, a group of 15 hits shared a common catechol structural moiety. The binding model of Hit 1 suggested that this catechol portion occupied the pTyr-binding pocket. The catechol mimicked pTyr by hydrogen bonding to Arg609 and Glu612, which are conserved cross most SH2 domains. In biochemical assays, 1 inhibited STAT3 DNA-binding and competed with pTyr peptides in binding to the STAT3 SH2 domain. This suggests that the catechol moiety may be a pTyr bioisostere and might be generally used for designing novel cell-permeable SH2 inhibitors.

10:45 60 Searching for new targets for the inhibition of Acetyl-CoA Carboxylase
Matt E. McKenzie, mckenzie17@cox.net1, Grover L. Waldrop, N/A2, and Bin Chen, binchen@lsu.edu1. (1) Department of Chemistry, Louisiana State University, Choppin 400, Baton Rouge, LA 70802, (2) Department of Biological Sciences, Louisiana State University, Baton Rouge, LA 70803

Current research on the inhibition Acetyl-CoA carboxylase(ACCase) could lead to novel antiobesity and antimicrobial agents. Bacterial ACCases contain three different subunits, a central biotin carrier protein which transfers the biotin between the biotin carboxylase subunit and carboxyltransferase subunit. We used the ZINC database and the AutoDock program to screen for potential targets on both of these subunits. Since biotin is the natural ligand in both of these subunits, similar top dockers and binding motifs are found. These top docked ligands are checked with an experimental inhibition study. This presentation will explore the similarities between the tight binding ligands of these two subunits and the reasons why certain ligands showed no experimental inhibition.

11:10 61 Computer-aided design of novel Akt inhibitors targeting to the pleckstrin homology domain
Lei Du-Cuny, lducuny@mdanderson.org1, Garth Powis1, Emmanuelle J. Meuillet2, Eugene Mash3, and Shuxing Zhang1. (1) Department of Experimental Therapeutics, MD Anderson Cancer Center, 1515 Holcombe Blvd. - Unit 36, Houston, TX 77030, (2) Departments of Nutritional Sciences and Molecular and Cellular Biology, College of Agriculture and Life Sciences, 1177 E. 4th Street, PO BOX 210038, Tucson, AZ 85721-0038, (3) Department of Chemistry, The University of Arizona, 1306 E University, Tucson, AZ 85721

PI3K/Akt pathway represents a new cancer target for drug discovery. We have employed multiple molecular docking approaches to design and study non-lipid based inhibitors targeting to Akt PH domain. Guided by computational modeling, several hits have been identified and their anticancer activities were experimentally tested. The most active compound exhibits binding affinity as high as 25 nM. However, in silico ADMET studies were conducted and demonstrated that compounds with nitrobenzyl moiety were possibly toxic, consistent to known facts that they are usually mutagens and carcinogens. Based on docking results, we have proposed a series of analogues in which the nitrobenzyl groups were replaced but protein-ligand interaction patterns were maintained. Further lead identification and optimization is in progress to improve the ADME/Tox properties of these compounds. In conclusion, molecular modeling has been successfully employed to guide our chemical design and optimization of novel Akt PH domain inhibitors.

MONDAY MORNING

CHED - Using Social Networking Tools to Teach Chemistry
Hilton New Orleans Riverside - Oak Alley
Organized by: Harry E. Pence, Laura E. Pence
Presiding: Laura E. Pence, Hilary J. Eppley
8:30   Introductory Remarks
8:35 184 Development of chemical forums, blogs, wikis, RSS feeds, crowdsourcing, and social bookmarking websites for chemistry
Mitch A. Garcia1, M. N. Ali2, Noel N Chang2, and Heino Nitsche, hnitsche@lbl.gov3. (1) College of Chemistry, UC Berkeley, 446 Latimer, Berkeley, CA 94720, (2) Nuclear Science Division, LBNL and Department of Chemistry, UC Berkeley, 1 Cyclotron Road, MS 88R0192, Berkeley, CA 94720, (3) Nuclear Science Division, University of California, Berkeley, Department of Chemistry and Lawrence Berkeley National Laboratory, 1 Cyclotron Road, MS 70R0319, Berkeley, CA 94720

Several social networking platforms have been created to facilitate chemical education in recent years by us. Chemicalforums.com is a community driven forum where any student can ask a question and any member can help answer. Chemblogs.org is a website for chemists to obtain their own blogs, similar to Blogger.com and Wordpress.com. Chemmunity.com is a website where all members help solve a chemical research question; it combines both wiki editing capabilities with a crowdsourcing approach to solving chemical phenomenon. Further experiments mixing Web 2.0 technologies with chemistry will be discussed.

8:55 185 Using Facebook as an online scientific community where students can interact with each other and discuss chemical concepts covered in the laboratory
Jacob D Schroeder, jds4097@iastate.edu, Department of Chemistry, Iowa State University of Science & Technology, 3051 Gilman Hall, Ames, IA 50011-3111, Thomas J. Greenbowe, tgreenbo@iastate.edu, Department of Chemistry, Iowa State University of Science and Technology, 3051 Gilman Hall, Ames, IA 50011, and Gerry McKiernan, gerrymck@iastate.edu, Library, Iowa State University, 152 Parks, Ames, IA 50011

Online social networks such as Facebook have significantly grown in user volume across the world. The Iowa State University Facebook network is composed of nearly 35,000 people alone. Despite these numbers and growth in online connectivity, we have noticed a corresponding lack of user traffic with the WebCT® discussion forum that we had been using. In response to this decreasing level of usage, we have collaborated with our Science & Technology librarian to develop a group page on Facebook dedicated to our one semester undergraduate organic chemistry course for non-science majors. This presentation will give an overview of how Facebook has been used for the students in the course and what further could be accomplished with it. It will also focus on the differences between the two online communities, user volume, and the level of discussion that takes place between students, as well as between students and the instructor.

9:15 186 Being there: Using social networking services for engaged library instruction
Gerry McKiernan, gerrymck@iastate.edu, Library, Iowa State University, 152 Parks, Ames, IA 50011

At colleges and universities today, a significant portion of students are members of Facebook (http://www.facebook.com), the online social networking service. Beginning in Summer 2007, we initiated a series of outreach projects using Facebook to directly promote Library programs and services to select members of the ISU community. These initiatives sought to inform students and faculty not only about the availability of core services offered by the reference and instruction department (e.g., book and journal selection, library presentations, research assistance) but also about the general library services provided by the library (e.g., interlibrary loan, library collections, reserve and media services).

This presentation will report on the results of these projects and describe future outreach plans. In addition, the general nature of Facebook, and its current and potential use by educators will be reviewed. The presentation will conclude with a brief overview of key readings and Web resources about online social networks.

9:35   Intermission
9:45 187 Off the beaten path: del.icio.us and social tagging in the classroom
Laura E. Pence, LPence@hartford.edu, Department of Chemistry, University of Hartford, 200 Bloomfield Ave, West Hartford, CT 06117

Web utilities such as del.icio.us, connetea, and sitesUlike offer powerful advantages over the default bookmarking capabilities of browsers because the web utilities are computer independent and allow websites to be tagged and organized according to keywords defined by the user. Tags and tagged sites may also be shared among users in a social network thus creating an expanded database of filtered and evaluated content resources. Incorporating del.icio.us into a course provides a variety of enrichment opportunities. Students can be taught valuable skills of assessing websites for bias and reliability, and the tagged sites expand students' exposure to current issues, which enhances classroom discussion. Details of del.icio.us projects in a non-science majors class and in an advanced chemistry class will be discussed

10:05 188 Social networking: The challenges of audience participation
Jason Wilde, Nature Publishing Group, 4 Crinan Street, London, N19XW, United Kingdom

Scientific communication has changed dramatically in recent years and the introduction of new web-based tools is making this flow of information simpler and faster.

This talk focuses on three new initiatives from Nature Publishing Group (NPG). Over the past year, NPG has launched a number of scientific blogs, some more popular than others. Whilst blogs are simple to create, it is important to know what you want to achieve and what your community needs. Journal Club highlights the best research, identified by our readers, and through audience participation, recommendations can be ranked and discussed. Nature Protocols is an interactive resource for laboratory protocols for bench researchers. Protocols are presented in a 'recipe' style providing step-by-step descriptions, which readers can immediately apply to their own research.

This presentation introduces each of these new projects and discusses the benefits and challenges of launching web tools that rely on user contributions.

10:25 189 Social networking and virtual worlds at the Nature Publishing Group
Joanna Scott, j.scott@nature.com, Web Publishing, Nature Publishing Group, 4 Crinan Street, London, N1 9XW, United Kingdom

2007 may be remembered as the year that using the internet for social networking really became mainstream. It was not limited to students and professional networkers, but has also had a profound impact on scientists and publishers. This looks set to be more than a flash in the pan and become a crucial means of communication.

This talk will focus on three very different projects from NPG, and look how they have succeeded or failed. Nature Network, a social networking site for scientists. Nature Precedings, our pre-print service for scientific papers, allowing researchers to share results with the community before publication, to get feedback and avoid delays in getting infomation out. Second Nature, in the virtual world Second Life, is our most unconventional communication platform, but has played host to a fascinating speaker series, and will expand in the new year into mini-conferences.

This talk will assess each project and discuss how they fit into the wider picture of social networking in research, education and publishing. We will outline the lessons learned so far, the next steps, and look at what the future of social networking might look like for the scientific community.

10:45   Intermission
10:55 190 Project Prospect from the RSC: The evolving journal article and chemical education
Richard Kidd, kiddr@rsc.org, Royal Society of Chemistry, Thomas Graham House, Science Park, Milton Road, Cambridge, CB4 0WF, United Kingdom

The Royal Society of Chemistry's Project Prospect has introduced semantic enrichment and the inclusion of structured science into primary research articles, and the project won the 2007 ALPSP/Charlesworth Award for Publishing Innovation. The development and future direction of this project will be discussed, demonstrating the possibilities for the use of article information within education and the possible evolution of the scientific article in the years to come.

11:15 191 Bridging the divide: The integration of scientific data and journal literature
Barbara Losoff, barbara.losoff@colorado.edu, Science Library, University of Colorado at Boulder, 184 UCB, Norlin Library, Boulder, CO 80309-0184

The sciences are disciplines largely driven by facts or data (Seringhaus and Gerstein 2007). Historically scientific databases and scholarly articles have operated as discreet units, with databases mostly seen as providing ‘background' to the journals ‘foreground'. The advent of large-scale digital repositories along with the need for sharing useful data world-wide, demands change to the current information structure. The integration of digital scientific data with scholarly literature has the potential to actualize semantic web design principles, creating “a universal medium for data, information, and knowledge exchange” (Wikipedia). This presentation will review the current literature; identify factors leading to integration, and discuss the future format of journals and databases and how such a transformation will ultimately benefit scientific research.

11:35 192 Open access: Toward the mainstream?
Bryan A Vickery, bryan.vickery@chemistrycentral.com and Gino G D'Oca, gino.doca@chemistrycentral.com. Chemistry Central, Middlesex House, 34 - 42 Cleveland Street, London, W1T 4LB, United Kingdom

There has been significant growth in the adoption of open access publishing since the first OA journals began publishing in 2000. The DOAJ lists almost 3000 such titles now, but just 65 of these are listed as "chemistry". Our estimate is that 10% of articles in PubMed will soon be freely available immediately after publication. This presentation will look at the spread of open access and open science from biomedicine to other disciplines, such as chemistry, and the benefits to researchers and educators. Models for funding open access publishing, the role of funding agencies and institutional repositories will also be discussed.

MONDAY AFTERNOON

PRES - Energy Research: Future Challenges and Opportunities
Morial Convention Center La Louisiane, Blrm. C
Jeffrey J. Siirola, Michelle Buchanan, Organizers
1:30   Welcome Remarks. Bruce E. Bursten, ACS President; Dale Keairns, AIChE President
1:40 64 Introduction and overview of the Basic Energy Science (BES) program
Patricia M. Dehmer
2:10 65 Assuring a secure energy future
Raymond L. Orbach, ray.orbach@science.doe.gov, Under Secretary for Sciences, U. S. Department of Energy, 1000 Independence Ave., S.W, Washington, DC 20585

Satisfying burgeoning global energy demand and curtailing greenhouse gas emissions in the century ahead will require more than incremental improvements in current technologies. To tackle these problems, the world will need transformational breakthroughs in basic science, leading to game-changing, “disruptive” technologies that fundamentally alter the way we produce, store, transmit, and use energy. Many of these new energy technologies are likely to emerge as an outgrowth of science's increasing ability to direct and control matter down to the molecular, atomic, and quantum levels. To accelerate this fundamental research, the Department of Energy's Office of Science has announced an Energy Frontiers Research Centers (EFRCs) initiative and includes $100 million under the President's Fiscal Year 2009 Budget Request to establish EFRCs (at $2-$5 million per year) around the nation. Universities, national laboratories, private companies, and nonprofit organizations will be invited to compete. Research areas will include solar energy utilization; superconductivity; solid-state lighting; advanced nuclear energy systems; combustion of 21st century transportation fuels; the hydrogen economy; electrical-energy storage; geosciences as it relates to the long-term storage of both carbon dioxide and spent nuclear fuel; materials under extreme environments; and catalysis for energy-related processes.

2:40 68 Hydrogen economy
Mildred Dresselhaus, millie@mgm.mit.edu, Institute Professor of Physics & Electrical Engineering, Massachusetts Institute of Technology, Room 13-3005, 77 Massachusetts Ave., Cambridge, MA 02139

One of the Grand Challenges of the 21st Century is to achieve a sustain-able energy supply as the global demand per capita of energy consumption increases sharply, fossil fuel supplies decline and environmental concerns mount. In this talk the role that hydrogen might play in addressing the Grand Energy Challenge is considered, including hydrogen production, storage and utilization, with emphasis given to the large gap between present science and technology know-how and the requirements in efficiency and cost for hydrogen to play its expected role. Opportunities for nanoscience and nanotechnology to narrow this gap will be discussed, and examples of recent progress will be presented.

3:00 67 Advanced nuclear energy systems
James B. Roberto, robertojb@ornl.gov, Deputy Director for Science and Technology, Oak Ridge National Laboratory, Oak Ridge National Laboratory, PO BOX 2008 MS6240, Oak Ridge, TN 37831-6240

The projected growth in nuclear power has focused increased attention on identifying priority research directions that underpin the development of advanced materials, fuels, waste forms, and separations technologies for the effective utilization of nuclear energy. In all of these areas, the performance of materials and chemical processes under extreme conditions is a limiting factor. The fundamental challenge is to understand and control chemical and physical phenomena in multi-component systems from femtoseconds to millennia, at temperatures to 1000 centigrade, and radiation doses to hundreds of displacements per atom. Addressing this challenge provides an opportunity to revolutionize the science and technology of advanced nuclear energy systems by enabling new materials, chemical processing, and predictive modeling. This presentation will draw from the recommendations of the Office of Basic Energy Sciences Workshop on Basic Research Needs for Advanced Nuclear Energy Systems, U.S. Department of Energy, October 2006, www.sc.doe.gov/bes/reports/files/ANES_rpt.pdf.

3:20 69 Electrical energy storage
Héctor D. Abruña, hda1@cornell.edu, Department of Chemistry and Chemical Biology, Cornell University, Baker Laboratory, Ithaca, NY 14853-1301

This presentation will cover basic research needs and opportunities underlying batteries, capacitors and related technologies, with a focus on new or emerging science challenges with potential for significant long-term impact on the efficient storage and release of electrical energy. Highlighted areas will include coupled ionic and charge transport, electrolyte physics, theory and modeling, and novel materials and approaches.

3:40 70 Catalysis for transportation fuels
Bruce C. Gates, bcgates@ucdavis.edu, Distinguished Professor, Department of Chemical Engineering and Materials Science, University of California, One Shields Avenue, Davis, CA 95616

Catalysis is the essential technology for production of transportation fuels from petroleum and natural gas. As heavier fossil feedstocks, including coal, continue to replace light petroleum, they pose major research challenges, because these feedstocks are increasingly complex in composition and structure and higher in contaminant sulfur and nitrogen. Fundamental advances are required in feedstock and product analysis, elucidation of reactant-catalyst interactions, and the application of theory with experiment to guide catalyst discovery. Successful addition of biomass to the feedstock pool will require extensive research along similar lines focused on the discovery of new classes of catalysts, specifically to facilitate the removal of oxygen and increase hydrogen-to-carbon ratios.

4:00 66 Solar energy utilization
Nathan S. Lewis, nslewis@its.caltech.edu, George L. Argyros Professor Chemistry, Division of Chemistry and Chemical Engineering, California Institute of Technology, 1200 E. California Blvd., Pasadena, CA 91125

Advances in nanoscience and nanotechnology are enabling dramatic improvements in our ability to cost-effectively use solar energy. There are three approaches in general to solar energy utilization: solar electricity, solar fuel, and solar thermal technologies. Each has its own technology bottlenecks and hurdles that are being addressed through advances in nanoscience and nanotechnology. The ability to orient matter on the nanoscale has led to new physics of charge carrier collection in solids for solar energy conversion to electricity as well as the promise of cheap paintable solar cells. The ability to mimic nature is leading to artificial photosynthetic systems that directly make fuel from the sun with efficiencies competitive to or higher than that of plants and bacteria. Finally the ability to effectively utilize different wavelengths of the solar spectrum in an integrated fashion in high efficiency solar cells and thermoelectrics is making possible significant advances in the efficiency of solar thermal systems. These approaches and areas where science and technology can have broad impact will be discussed in this presentation.

TUESDAY MORNING

COMP - Drug Discovery: Mostly about Ligands
Ernest N. Morial Convention Center 348
I. Visiers, Organizer; E. Fayfant, Presiding
8:00 93 Applications of target class pharmacophore fingerprint modeling and multi-objective genetic algorithm optimization to large-scale combinatorial library design for corporate compound collection enhancement
G. Patrick Brady, Pat.G.Brady@gsk.com and Zheng P. Yang, Zheng.P.Yang@gsk.com. Computational and Structural Sciences, GlaxoSmithKline, 1250 South Collegeville Road, UP12-210, P. O. Box - 5089, Collegeville, PA 19426

This presentation describes a GSK in-house pharmacophore fingerprint (pFP) toolkit and lead ID model building tool, pFPbitrank. Fundamental issues relating to pFP descriptors will be addressed in the talk- including lead hopping capacity- and a detailed description of pFPbitrank will be provided. pFPbitrank builds a model by identifying the 3- or 4-point pharmacophores (i.e. pFP bits) that preferentially discriminate a training set of active compounds from inactives. Randomization trials are conducted to assess the level of statistical signal amongst the most informative bits and to decide how many of these bits should comprise the pFP model. The pFPbitrank methodology will be applied to both single-target and multi-target activity data and the resulting models will be used to perform virtual screens; enrichment will be evaluated via ROC curves. Particular attention will be paid to a 7TM target class model, which has been used extensively at GSK to conduct 7TM-focused enhancement of the corporate screening collection via external compound acquisition and the design of large combinatorial libraries.

8:25 94 Applications of target class pharmacophore fingerprint modeling and multi-objective genetic algorithm optimization to large-scale combinatorial library design for corporate compound collection enhancement
Zheng Yang, Zheng.P.Yang@gsk.com, Computational and Structural Chemistry, Molecular Discovery Research, GlaxoSmithKline Pharmaceuticals, 1250 South Collegeville Road, Collegeville, PA 19426 and G. Patrick Brady, Pat.G.Brady@gsk.com, Computational and Structural Sciences, GlaxoSmithKline, 1250 South Collegeville Road, UP12-210, P. O. Box - 5089, Collegeville, PA 19426

Corporate HTS collections in the pharmaceutical industry are a patchwork of sets of compounds that were either made/acquired for a particular molecular target or were designed to cover chemistry space as completely and efficiently as possible. The former approach produces compound sets of great “depth” – i.e. compounds densely clustered in particular chemotypes, such as those identified via focused virtual screening or pursued via lead optimization efforts. The latter approach produces compound sets of great “breadth” – i.e. that cover chemical space very thinly via large numbers of small chemotype clusters. Historically, little collection enhancement effort has been spent in between these extremes of depth and breadth. This presentation will discuss applications of target class pharmacophore fingerprint (pFP) modeling1 and multi-objective genetic algorithm (MoGa) optimization2 to the design of large combinatorial libraries that begin to fill this gap. Advantages of target class focused MoGa library design over pure compound collection enhancement-based MoGa library design will be presented in a detailed case study of a large 7TM-focused array design. The impact of target class focused MoGa library design on the GSK HTS collection will also be discussed.

1. Brady, P.G., Ligand-based Design at GSK via pFPs, 232nd American Chemical Society National Meeting, San Francisco, CA, USA., September 10-14, 2006.

2. Gillet, V. J.; Khatib, W.; Willett, P.; Fleming, P. J.; Green, D. V. S. Combinatorial Library Design Using a MultiobjectiveGenetic Algorithm. J. Chem. Info. Comput. Sci. (2002), 42(2), 375-385

8:50 95 Confirm: Connecting fragments in receptor molecules
David C. Thompson1, Aldrin Denny, Diane Joseph-McCarthy, DJoseph@wyeth.com2, Christine Humblet1, and Eric Feyfant3. (1) Chemical and Screening Sciences, Wyeth Research, 200 CambridgePark Drive, Cambridge, MA 02140, (2) Department of Structural Biology & Computational Chemistry, Wyeth Research, Chemical & Screening Sciences, 200 Cambridge Park Drive, Cambridge, MA 02140, (3) Department of Chemical and Screening Sciences, Wyeth Research, 200 CambridgePark Dr, Cambridge, MA 02140

A novel algorithm for the connecting of fragment molecules is presented and validated for a number of test systems. Within the CONFIRM (Connecting Fragments in Receptor Molecules) approach a pre-prepared library of bridges is searched to extract those which match a search criterion derived from known experimental or computational binding information about fragment molecules within a target binding site. The resulting bridge ‘hits' are then connected, in an automated fashion, to the fragments and docked into the target receptor. Docking poses are assessed in terms of root-mean-squared deviation from the known positions of the fragment molecules, as well as docking score should known inhibitors be available. The creation of the bridge library, the full details and novelty of the CONFIRM algorithm, and the general applicability of this approach within the field of fragment-based de novo drug design are discussed.

9:15 96 Fast and accurate method for flexible ligand superposition and shape-based screening
Steven L. Dixon, dixon@schrodinger.com1, Pranav Dalal, dalal@schrodinger.com2, Jas Gata-Aura1, Shashidhar N Rao, rao@schrodinger.com1, John C. Shelley, jshelley@schrodinger.com2, and B. Woody Sherman, Woody.Sherman@schrodinger.com1. (1) Schrödinger, Inc, 120 West Forty-Fifth Street, 32nd Floor, Tower 45, New York, NY 10036-4041, (2) Schrodinger, Inc, 3rd Floor, G. Pulla Reddy Building, 6-3-879 & 879/B, Begumpet, Hyderabad, India

The use of 3-dimensional molecular shapes has been demonstrated to be useful in comparing small molecules. We have developed a novel method that allows for rapid superposition and scoring of flexible molecules. The core algorithm is based on the alignment of a set of optimal atom triads followed by volume overlap scoring. The method can process approximately 1000 conformations per second on a modern computer. For flexible superposition we demonstrate on a large data set of crystal structures how accurate alignments can be obtained rapidly (less than one second per molecule) in an automated fashion. Next, we apply the method to virtual screening and show high enrichment rates across a broad range of targets and ligands. A one million compound database (100 conformations per compound) can be processed in approximately 15 minutes on a 100-processor cluster, making this method attractive for pre-screening large databases before downstream pharmacophore-based or docking screens.

9:40   Intermission
9:55 97 Combining clique-detection, MOGUL and MOGA for pharmacophore generation
David A. Cosgrove, David.Cosgrove@astrazeneca.com1, Eleanor J. Gardiner, e.gardiner@sheffield.ac.uk2, Valerie J. Gillet, v.gillet@sheffield.ac.uk2, and Robin Taylor, robin.t@virgin.net3. (1) AstraZeneca, Mereside, Alderley Park, Macclesfield, United Kingdom, (2) Department of Information Studies, University of Sheffield, Western Bank, Sheffield, United Kingdom, (3) 54 Sherfield Avenue, Rickmansworth WD3 1NL, UK, United Kingdom

Pharmacophore elucidation is a difficult problem involving the determination of the 3D description of interactions between a small molecule and a protein without knowledge of the protein structure. A Multi-Objective Genetic Algorithm (MOGA) has been developed with the aim of generating multiple feasible solutions (Cottrell et al, JCAMD,20,735-749,2006). However the solution space of potential pharmacophores is very large and increases with the number of molecules. In this work we have combined a clique-detection algorithm with the MOGA in order to limit the MOGA exploration to a feasible reason of solution space to increase both the efficiency and effectiveness of the program. In a further enhancement we bias the search towards reasonable conformers using MOGUL (Bruno et al., J. Chem. Inf. Comput. Sci., 44,2133-2144,2004). We report the results of these enhancements in terms of both speed and solution quality on datasets of up to ten molecules.

10:20 98 QSAR-based design of novel anti-HRV 2 agents
Anatoly Artemenko, artanat@ukr.net1, Eugene N. Muratov, murik@ccmsi.us1, Victor E. Kuz'min, victor@2good.org2, Ekaterina Varlamova, 00dqsar@ukr.net3, Vadim Makarov4, Olga Riabova4, Michaela Schmidtke, Michaela.Schmidtke@med.uni-jena.de5, and Peter Wutzler5. (1) Computational Center for Molecular Structure and Interactions, Jackson State University, 1400 J.R. Lynch Str, Jackson, MS 39217, (2) Laboratory of Theoretical Chemistry, A.V.Bogatsky Physical-Chemical Institute NAS of Ukraine, Lustdorfskaya Doroga 86, Odessa, 65080, Ukraine, (3) Chemical-Technological Department, Odessa National Polytechnical University, 1 Shevchenko Av, Odessa, 65000, Ukraine, (4) Research Center for Antibiotics, 3A Nagatinskaya Str, Moscow, Russia, (5) Institute of Virology and Antiviral Therapy, Friedrich Schiller University, Bachstraße 18, Jena, 07743

The objectives of the present study were development of QSAR analysis of antiviral activity of a set of [(biphenyloxy)propyl]isoxazole derivatives that inhibit human rhinovirus 2 (HRV-2) replication and design of novel potential antiviral agents on the base of obtained results.

The QSAR approaches applied were simplex representation of molecular structure (SiRMS) and Lattice Model (LM). The relationship between antiviral activity against the HRV-2, cytotoxicity in HeLa cells, selectivity index and structure of [(biphenyloxy)propyl]isoxazole derivatives has been studied systematically.

Quite adequate QSAR models have been obtained using PLS method and have been used as consequent virtual screening tool. Structural fragments with positive or negative influence on cytotoxicity as well as antiviral activity have been determined on the base of these models. This information has been used for directed drug design of novel antirhinoviral agents. High level of antiviral action and selectivity of several designed compounds has been verified experimentally.

WEDNESDAY MORNING

COMP - Drug Discovery: Drug Discovery
Ernest N. Morial Convention Center 348
I. Visiers, Organizer; P. Carlqvist, Presiding
8:00 224 Binding response: A method for the prediction of ligand binding sites on proteins
Shijun Zhong, sjzhong@gmail.com, Department of Pharmaceutical Sciences, University of Maryland, 20 Penn Street, Baltimore, MD 21201 and Alexander D. MacKerell Jr., alex@outerbanks.umaryland.edu, Department of Pharmaceutical Sciences, School of Pharmacy, University of Maryland, 20 Penn St, Baltimore, MD 21201

Binding response is a novel descriptor to evaluate the response of a putative binding site on a protein to probe compounds using both the geometry and energies of binding poses. A complete approach has been implemented including the generation of the protein surface, identification of putative binding sites, docking of a set of 1000 diverse compounds and evaluation of the putative binding sites using the binding response. This method is proposed to facilitate the identification of binding sites on protein surfaces for use in computational database screening studies targeting protein-ligand and protein-protein interactions. Analysis of 29 protein-ligand complexes shows a 90% success rate in identifying known binding sites.

8:25 225 Giving the Rule-of-5 a more accurate twist
Greg Pearl, greg.pearl@acdlabs.com, Sanjivanjit Bhal, sanji.bhal@acdlabs.com, Ian G. Peirson, and Karim Kassam. Advanced Chemistry Development, Inc, 110 Yonge Street, 14th Floor, Toronto, ON M5C 1T4, Canada

The much publicized “Rule-of-5” (Ro5) has been widely adopted in the pharmaceutical industry as the first step in the virtual screening of compound libraries, in an effort to pre-emptively eliminate hits that are deemed to have poor physicochemical properties for oral bioavailability.

LogP is a key parameter in the Ro5 and, although useful, fails to take into account variation in drug lipophilicity due to ionization under physiological conditions. Given that more than 95% of commercial pharmaceuticals contain an ionizable moiety, we propose that logD is a better descriptor for lipophilicity in the Ro5 (and similar filters). This alternative value should help reduce the number of potential false-positives eliminated in screening.

In this presentation we will discuss results from screening a number of commercially-available libraries using an adapted Ro5 applying logD in place of logP.

8:50 226 Virtual screening for superior R-groups
Richard D. Cramer, cramer@tripos.com, Tripos, 1699 South Hanley Road, St. Louis, MO 63144

Frequently during lead optimization, only a side chain or two are amenable to further structural variation, while established CADD methods are weak for choosing among even the limited number of candidates first coming to mind. Topomers provide an alternative possibility of "virtual screening for R-groups" within the largest structural databases, and on the basis of 3D-QSAR potency predictions as well as shape similarity. However that possibility is so far untested. We will therefore report retrospective studies to address at least the following questions, starting from 25 reported topomer CoMFA models that statistically replicate published 3D-QSAR studies:

Within the ZINC database, how frequently do there indeed exist otherwise reasonable R-groups having higher predicted potencies?

How reliable are such predictions of higher potency? (How often do topomer CoMFA analyses omitting the most potent known structures correctly flag such omitted structures as promising? And what caveats are associated with such predictions?)

9:15 227 Histone deacetylase inhibitors: Reasons for isoform selectivity
Guillermina. Estiu, gestiu@nd.edu1, Olaf Wiest, owiest@nd.edu1, Edward Greenberg2, Ralph Mazitschek, ralph@broad.harvard.edu2, and James Bradner, James_Bradner@dfci.harvard.edu3. (1) Department of Chemistry and Biochemistry, University of Notre Dame, 251 Nieuwland Science Hall, Notre Dame, IN 46556, (2) Broad Institute of Harvard University and MIT, Cambridge, MA 02142, (3) Dana-Farber Cancer Institute, 44 Binney Street, Boston, MA 02115

Histone deacetylases (HDACs) are promising targets in drug development, regulating gene expression through the deacetylation of histone tails. Eukaryotic HDACs have been classified into four groups on the basis of a phylogenetic analysis.

A wide range of structures have been identified that are able to inhibit the activity of the different classes, achieving significant biological effects in preclinical models of cancer. However, only few molecules are emerging as preferential inhibitors of class 1 versus class 2, and even fewer are able to discriminate efficiently among HDACs belonging to the same class. Nevertheless, common patterns, such as a low activity in HDAC8, have emerged.

Using x-ray structures, homology modeling, docking, and long-timescale MD simulations, possible reasons for the isoform selectivity are elucidated. The contributions of interactions in the binding, linker, and cap regions of the inhibitors to the potency and selectivity of a range of HDAC inhibitors are discussed

9:40 228 Modeling the metabolic space in drug discovery
Lothar Terfloth, terfloth@molecular-networks.com1, Bruno Bienfait Bienfait, bruno.bienfait@chemie.uni-erlangen.de2, Johann Gasteiger, Gasteiger@ccc.chemie.uni-erlangen.de3, and Christof H. Schwab, schwab@molecular-networks.com1. (1) Molecular Networks GmbH, Henkestrasse 91, D-91052 Erlangen, Germany, (2) University of Erlangen-Nuremberg, Naegelsbachstrasse 25, 91052, Germany, (3) Computer-Chemie-Centrum, University of Erlangen-Nurnberg, Nagelsbachstrasse 25, Erlangen, D-91052, Germany

In silico prediction of ADMET (absorption, distribution, metabolism, elimination, toxicity) properties is of special interest in the drug discovery process in order to detect and eliminate compounds with inappropriate pharmacokinetic properties at an early stage. A central step in the ADMET profiling of potential drug candidates is the assessment of drug metabolism. Some enzymes involved in the detoxification process show polymorphism and have multimodal binding sites. The majority of the oxidation reactions in phase I metabolism are catalyzed by cytochrome P450 enzymes.

This paper focuses on several aspects related to phase I metabolism by cytochrome P450 enzymes. Models for the prediction of the isoform specificity of cytochrome P450 3A4, 2D6, and 2C9 substrates [1] as well as for the prediction of potential inhibition of cytochrome P450 2D6 and 2C9 will be presented. The impact of descriptor selection, the choice of the model building method and the selection of training and test data set will be carefully discussed. A comprehensive scheme of cross-validation experiments was applied to assess the robustness and reliability of the models developed. In addition, the predictive power was inspected by predicting an external validation data set. The final models that perform with a predictability of over 80% are implemented in the program system isoCYP.

[1] Terfloth, L.; Bienfait, B.; Gasteiger, J. Ligand-Based Models for the Isoform Specificity of Cytochrome P450 3A4, 2D6, and 2C9 Substrates. J. Chem. Inf. Model.2007, 47, 1688-1701.

[2] isoCYP is available from Molecular Networks GmbH, Erlangen, Germany and available for testing at www.molecular-networks.com/online_demos/cyp450.

WEDNESDAY AFTERNOON

COMP - Model Applicability Domains: When Can I Use my Model?
Ernest N. Morial Convention Center 347
C. M. Breneman, D. F. Ortwine, Organizers
1:00   Introductory Remarks
1:10 235 Domain applicability of ligand and structure-based virtual screening
Anthony Nicholls1, Mark McGann, mcgann@eyesopen.com2, and Paul Hawkins, phawkins@eyesopen.com2. (1) OpenEye Scientific Software, Inc, 9d Bisbee Court, Santa Fe, NM 87508, (2) OpenEye Scientific Software, 222 3rd Street Suite 3211, Cambridge, MA 02142

We present results on the virtual screening behavior of docking and shape-based ligand methods as applied to large test sets such as DUD. In addition to lessons on the applicability of methods to particular systems, we demonstrate significant synergy between methods with substantial independence of domain applicability, some limits on the expected behavior of systems that are similar but not identical, and some suggested best practices for virtual screening.

1:35 236 Testing the limits of a QSAR model: How many cases are actually needed to develop a reliable predictive model?
C. Matthew Sundling, sundlm@rpi.edu1, Curt M. Breneman, brenec@rpi.edu2, Mark J. Embrechts3, Changjian Huang, huangc@rpi.edu4, Xiaohua Wu3, and N. Sukumar, nagams@rpi.edu5. (1) Department of Chemistry and Center for Biotechnology and Interdisciplinary Studies, Rensselaer Polytechnic Institute, 110 8th Street, Troy, NY 12180-3590, (2) Department of Chemistry / RECCR Center, Rensselaer Polytechnic Institute, 110-8th Street, Center for Biotechnology and Interdisciplinary Studies, Troy, NY 12180, (3) Department of Decision Sciences & Engineering Systems, Rensselaer Polytechnic Institute, 110 8th Street, Troy, NY 12180, (4) RECCR Center, Rensselaer Polytechnic Institute, 110-8th Street, Center for Biotechnology and Interdisciplinary Studies, Troy, NY 12180, (5) Department of Chemistry and Center for Biotechnology and Interdisciplinary Studies, Rensselaer Polytechnic Institute / RECCR Center, 110 8th St., Troy, NY 12180-3590

During this talk, key indicators of the predictive power of a statistical learning model will be presented as a function of the number and distribution of molecules in the training set, and the descriptors and learning method used. The study was based on the development of QSAR models for several representative datasets where the number and distribution density of the training data was gradually changed. The modeling performance of the remaining data was then evaluated for all descriptors, as well as for for selected subsets of descriptors with different levels of information content. Molecular features used in the study included MOE 2D and i3D descriptors as well as TAE / RECON, wavelet and PEST shape/property hybrid descriptors.

2:00 237 Automatic detection of outliers prior to QSAR studies
Alexander Golbraikh, golbraik@email.unc.edu, Hao Zhu, Lin Ye, lye@email.unc.edu, Mei Wang-Bell, meiwang@email.unc.edu, Hao Tang, tangh@email.unc.edu, and Alexander Tropsha, tropsha@email.unc.edu. Laboratory for Molecular Modeling, University of North Carolina at Chapel Hill, School of Pharmacy, Chapel Hill, NC 27599

We have developed automatic procedures for detection and elimination of outliers prior to QSAR studies. Two types of outliers are distinguished: leverage outliers and activity outliers. The former are compounds dissimilar from all other compounds in the dataset. They can be found using a sphere-exclusion algorithm. In contrast, activity outliers are compounds that are similar to some compounds in the dataset, but their activities are significantly different from those of their nearest neighbors. These compounds are of particular concern in QSAR studies: e.g., they can form so-called Maggiora's cliffs (Maggiora G.M., J Chem Inf Model. 2006, 46, 1535) in the descriptor space. We will discuss the automatic detection and elimination of outliers as applied to several datasets. We show that taking into account activity outliers affords the significant increase in predictive power of QSAR models.

2:25   Intermission
2:40 238 Combining global and local approaches to model domain applicability
Rajarshi Guha, rajarshi.guha@gmail.com, NIH Chemical Genomics Center, Room 3005, 9800 Medical Center Drive, Rockville, MD 20850 and David T. Stanton, stanton.dt@pg.com, Procter & Gamble, Miami Valley Innovation Center, 11810 East Miami River Road, Cincinnati, OH 45252

The goal of an in-silico predictive model is to predict a biological or physical property of a molecule based on its structural features. The model is built using a training set of molecules with the assumption that any new molecule to be predicted will have a degree of structural similarity with the training set. However, even though a new molecule may be structurally similar to the training set from the perspective of the model, it may still be predicted poorly because it incorporates some feature to which the model is blind. We describe a two-step approach to the quantification of model applicability. First, we consider the relationship between a new molecule and the structural features encoded in a model. This step is based on a determination of the point density of the region of model space that a new molecule is located in. The second step takes into account the differences between a new molecule and the training set in the context of a "global" chemistry space, represented by a structural key fingerprint. Our initial results indicate that a descriptor-weighted similarity between the new molecule and the model leads to a better measure of domain applicability. In addition we will present some initial results that highlight the importance of considering both model-specific and global aspects of domain applicability.

3:50 239 QSAR model stability: How much information is in the data?
Dominic Ryan, dominic.ryan@comcast.net1, Margaret McLellan2, and Curt M. Breneman, brenec@rpi.edu2. (1) DRI, 544 Newtown Road, Littleton, MA 01460, (2) Department of Chemistry / RECCR Center, Rensselaer Polytechnic Institute, 110-8th Street, Center for Biotechnology and Interdisciplinary Studies, Troy, NY 12180

The evolution of a good QSAR or QSPR predictor can require significant expertise and effort. Frequently this work produces a single final model or ensemble of models which are then applied to new problems. When the model hypersurface is explored and multiple nearly-equivalent models are generated, the model stability can be viewed as an indication of model robustness and applicability. Stability can be assessed by the impact of these variations on predicted rank ordering of cases. This in turn can be cast as a form of entropy. When multiple quasi-equivalent models produce a consistent rank ordering, the information content of the predictor set is high and the entropy is low. Conversely if multiple models lead to shuffling of rank ordering, then the Rank Order Entropy is high. The ROE metric can be used as a measure of the information power of a dataset. These concepts will be illustrated with several datasets and problem types.

3:30 240 Domain applicability: How far are ideal and reality?
Eugene N. Muratov, murik@ccmsi.us1, Victor E. Kuz'min, victor@2good.org2, and Anatoly G. Artemenko, artanat@ukr.net2. (1) Computational Center for Molecular Structure and Interactions, Jackson State University, 1400 J.R. Lynch Street, Jackson, MS 39217, (2) Laboratory of Theoretical Chemistry, A.V.Bogatsky Physical-Chemical Institute NAS of Ukraine, Lustdorfskaya Doroga 86, Odessa, 65080, Ukraine

Domain applicability (DA) estimation is absolutely necessary for increase of QSAR predictions reliability. However, the situations when molecules belonging to model DA are badly predicted and vice versa are widespread. Present work is devoted to analysis how close are existing and developing DA procedures to their ideal sense.

Three different approaches for DA estimation (developed by us DA ellipsoid, DA rectangle and popular leverage method) have been applied for lattice and simplex 2D and 3D QSAR models developed for several different QSAR/QSPR tasks. Certainly, elimination of compounds which aren't belonging to model DA from prediction set increases quality of prediction during virtual screening or soft drug design. However, at the same time it leads to narrowing of chemical space covered by new compounds and impedes transgressing the bounds of investigated activity limited by training set. More drastic drug design is more risky but it allowing getting much more dramatic results.

THURSDAY AFTERNOON

COMP - Model Applicability Domains: When Can I Use my Model?
Ernest N. Morial Convention Center 347
C. M. Breneman, D. F. Ortwine, Organizers
1:00   Introductory Remarks
1:05 268 Applicability domains, space coverage, and predictive power of QSAR models
Alexander Tropsha, alex_tropsha@unc.edu1, Alexander Golbraikh, golbraik@email.unc.edu2, and Hao Zhu1. (1) Laboratory for Molecular Modeling, School of Pharmacy, University of North Carolina at Chapel Hill, CB # 7360, Beard Hall, School of Pharmacy, Chapel Hill, NC 27599-7360, (2) School of Pharmacy, University of North Carolina, CB # 7360, Beard Hall, School of Pharmacy, Chapel Hill, NC 27599-7360

For many years, our group has advocated the use of applicability domains (AD) as a mandatory component of QSAR modeling. The AD defines the boundaries of the chemistry space within which the models could be used reliably for predicting compounds' bioactivities. The specific definitions of AD depend on the descriptor types, similarity metrics, and particular modeling techniques. I shall discuss specific examples of AD implemented in our group in the context of k Nearest Neighbor (kNN) and Support Vector Machines (SVM) QSAR approaches as well as in special scoring function for protein-ligand docking. Expanding the AD naturally leads to higher coverage of the chemistry space; however it may also lead to lower prediction accuracy. Thus, I shall emphasize the challenges in finding optimal AD that afford a reasonable balance between chemical space coverage and prediction accuracy of QSAR models. I will also discuss the use of AD for outlier detection.

1:30 269 Testing the validity range of QSAR models using one-class support vector machines
Mark J. Embrechts, Department of Decision Sciences & Engineering Systems, Rensselaer Polytechnic Institute, 110 8th Street, Troy, NY 12180, Curt M. Breneman, brenec@rpi.edu, Department of Chemistry / RECCR Center, Rensselaer Polytechnic Institute, 110-8th Street, Center for Biotechnology and Interdisciplinary Studies, Troy, NY 12180, Changjian Huang, huangc@rpi.edu, RECCR Center, Rensselaer Polytechnic Institute, 110-8th Street, Center for Biotechnology and Interdisciplinary Studies, Troy, NY 12180, and N. Sukumar, nagams@rpi.edu, Department of Chemistry and Center for Biotechnology and Interdisciplinary Studies, Rensselaer Polytechnic Institute / RECCR Center, 110 8th St., Troy, NY 12180-3590

One-class Support Vector Machines (SVM) are a powerful method for outlier detection and anomaly detection. This presentation highlights the use of one-class SVMs for evaluating QSAR model validity. For example, one typical HIV RT inhibitor dataset contains at least five different molecular classes. In this case, five different QSAR models were built using PLS and K-PLS, each time leaving out all the molecules for a particular class. A one-class SVM was then applied to confirm that the activity of molecules from the remaining class cannot be adequately predicted if no training samples from that class were included. The general use for one-class SVMs for evaluating model validity for QSAR datasets in general will then be discussed. Modeling features include MOE, TAE/RECON and PEST shape/property hybrid descriptors.

1:55 270 Assessment of additive/nonadditive effects in SAR: Implications in the drug discovery iterative process
Yogendra Patel, yogendra.patel@manchester.ac.uk1, Valerie J. Gillet, v.gillet@sheffield.ac.uk2, Peter Willet2, Joaquin Pastor3, Trevor Howe4, and Julen Oyarzabal, joyarzabal@cnio.es5. (1) Manchester Interdisciplinary Biocentre, University of Manchester, 131 Princess Street, Manchester, M1 7DN, United Kingdom, (2) Department of Information Studies, University of Sheffield, Regent Court, 211 Portobello Street, Sheffield, S1 4DP, United Kingdom, (3) Department Medicinal Chemistry, Johnson & Johnson Pharmaceutical R&D, Jarama, 75, Toledo, 45007, Spain, (4) Molecular Informatics, Johnson & Johnson Pharmaceutical Research & Development, Turnhoutseweg 30, 2340, Beerse, Belgium, (5) Molecular Informatics, Johnson & Johnson Pharmeaceutical R&D. Current address: Spanish National Cancer Research Centre (CNIO), Jarama 75, Toledo, 45007, Spain

Free-Wilson analysis of structure-activity-relationships is common practice in medicinal chemistry; and, nowadays, this additivity principle is becoming more prominent due to the popular modular approach for rational drug design: fragment-based drug design. In addition, additivity is implicit in most scoring functions used in structure-based design.

Our aims are to determine if we can estimate which effects, additive or non-additive, are taking place in near complete combinatorial libraries for the biological response(s) under study and then to determine the minimum attributes of a data set (size, distribution of properties etc) necessary to reach the same conclusion about additive/non-additive effects.

Eight near complete combinatorial libraries have been utilized and they show the complexity of protein-ligand interactions: additive, non-additive and partially additive data sets. Then, a series of retrospective experiments are carried out providing some guidelines on when it is appropriate to apply Free-Wilson analysis and implications for iterative drug design.

2:20   Intermission
2:35 271 Similarity based assessment of model applicability domain and quantitative evaluation of the reliability of the prediction
Pranas Japertas, jurgutis@ap-algorithms.com1, Andrius Sazonovas, andrius@pharma-algorithms.com1, Remigijus Didziapetris, remis@pharma-algorithms.com2, and Alanas Petrauskas, jurgutis@ap-algorithms.com3. (1) Faculty of Chemistry, Vilnius University, Naugarduko g. 24, Vilnius, LT-03225, Lithuania, (2) Pharma Algorithms, Inc, A.Mickeviciaus g. 29, LT-08117 Vilnius, Lithuania, (3) Pharma Algorithms Inc, 2700-161 Bay St. TD Tower, Toronto, ON M5J 2S1, Canada

Development of a methodology for the evaluation of Model Applicability Domain is presented using similarity analysis of the compounds in the training set. A novel methodology relying on the fact that any empirical in silico model works only for similar compounds was developed. The availability of similar compounds in the training set and experimental data consistency for such compounds was pivotal. This information is reflected in a corresponding Reliability Index (RI), which generates values from 0 (not reliable) to 1 (very reliable), assisting in interpretation of the results. The methodology is illustrated with examples of its application in estimating Model Applicability Domain for the models of logP, logD, solubility and toxicity. The reliability index is shown to be closely related to the overall quality of any given prediction that is represented by a clear correlation of the RI and RMSE values.

3:00 272 Localizing uncertainty in PLS predictivity
Robert D. Clark, bclark@tripos.com, Tripos International, 1699 S. Hanley Rd., St. Louis, MO 63144, Gunther Stahl, gstahl@tripos.com, Tripos, Inc, 1699 South Hanley Road, St. Louis, MO 63144, and Tamsin E. Mansley, tmansley@tripos.com, Tripos Informatics Research Center, 1699 S. Hanley Rd., St. Louis, MO 63144

Ordinary least squares regression (OLS) is convenient in that it provides a direct estimate of the uncertainty in prediction for any point in the descriptor space. Unfortunately, such estimates are only reliable when the underlying relationships are linear, the descriptors are mutually independent, and the observations themselves are identically and independently distributed (IID). Partial least squares with projection to latent structures (PLS) is typically applied instead when there are more descriptors than observations. It is also more appropriate when the descriptors are not independently distributed, however, as is generally the case for structure-activity relationships. Here, we will examine the effectiveness of distributing the predictive sum of squares (PRESS) from PLS across the individual observations in a training set to model the predictive uncertainty in different parts of the applicability domain.

3:25 273 Ensemble QSAR
George D. Purvis III, gpurvis@us.fujitsu.com, Biosciences, Fujitsu Computer Systems, 15244 NW Greenbrier Pkwy, Beaverton, OR 97007

Exhaustive searching for QSAR models through generation of all possible combinations of descriptors leads to dozens or even hundreds of models with r2 and errors very close to the best result but differing in one or more descriptors. The models are so close that it is difficult to justify one model over the other. We use this to advantage by creating ensembles of models and using the ensemble to estimate the errors in predicted results. When making predictions with the ensemble model, in addition to checking the chemicals predicted for compliance with the training set chemical space and training set descriptor and property ranges, we find that the distribution of predicted values from the ensemble of models is an excellent diagnostic of model applicability. In this talk we describe the method and present examples of its use.

 

 

Publishing software