Building a Chemical Informatics Grid Marlon Pierce Community

Building a Chemical Informatics Grid Marlon Pierce Community Grids Laboratory Indiana University

Acknowledgments n CICC researchers and developers who contributed to this presentation: q n n Prof. Geoffrey Fox, Prof. David Wild, Prof. Mookie Baik, Prof. Gary Wiggins, Dr. Jungkee Kim, Dr. Rajarshi Guha, Sima Patel, Smitha Ajay, Xiao Dong Thanks also to Prof. Peter Murray Rust and the WWMM group at Cambridge University More info: www. chembiogrid. org and www. chembiogrid. org/wiki.

Chemical Informatics and the Grid An overview of the basic problem and solution

Chemical Informatics as a Grid Application n n Chemical Informatics is the application of information technology to problems in chemistry. q Example problems: managing data in large scale drug discovery and molecular modeling Building Blocks: Chemical Informatics Resources: q Chemical databases maintained by various groups n q Application codes (both commercial and open source) n n q q n NIH Pub. Chem, NIH DTP Data mining, clustering Quantum chemistry and molecular modeling Visualization tools Web resources: journal articles, etc. A Chemical Informatics Grid will need to integrate these into a common, loosely coupled, distributed computing environment.

Problem: Connecting It Together n n The problem is defining an architecture for tying all of these pieces into a distributed computing system. q A “Grid” How can I combine application codes, web resources, and databases to solve a particular problem that interests me? q n Specifically, how do I build a runtime environment that can connect the distributed services I need to solve an interesting problem? For academic and government researchers, how can I do all of this in an open fashion? q Data and services can come from anywhere q That is, I must avoid proprietary infrastructure.

NIH Roadmap for Medical Research http: //nihroadmap. nih. gov/ n n The NIH recognizes chemical and biological information management as critical to medical research. Federally funded high throughput screening centers. q q q n 100 -200 HTS assays per year on small molecules. 100, 000’s of small molecules analyzed Data published, publicly available through NIH Pub. Chem online database. What do you do with all of this data?

High-Throughput Screening Testing perhaps millions of compounds in a corporate collection to see if any show activity against a certain disease protein

High-Throughput Screening n n n Traditionally, small numbers of compounds were tested for a particular project or therapeutic area About 10 years ago, technology developed that enabled large numbers of compounds to be assayed quickly High-throughput screening can now test 100, 000 compounds a day for activity against a protein target Maybe tens of thousands of these compounds will show some activity for the protein The chemist needs to intelligently select the 2 - 3 classes of compounds that show the most promise for being drugs to follow-up

Informatics Implications n n Need to be able to store chemical structure and biological data for millions of data points q Computational representation of 2 D structure Need to be able to organize thousands of active compounds into meaningful groups q Group similar structures together and relate to activity Need to learn as much information as possible (data mining) q Apply statistical methods to the structures and related information Need to use molecular modeling to gain direct chemical insight into reactions.

The Solution, Part I: Web Services n Web Services provide the means for wrapping databases, applications, web scavengers, etc, with programming interfaces. q q q n Many toolkits available q n WSDL definitions define how to write clients to talk with databases, applications, etc. Web Service messaging through SOAP Discovery services such as UDDI, MDS, and so on. Axis, . NET, g. SOAP, SOAP: : Lite, etc. Web Services can be combined with each other into workflows q q Workflow==use case scenario More about this later.

Basic Architectures: Servlets/CGI and Web Services Browser GUI Client Browser HTTP GET/POST Web Server WSDL SOAP Web Server JDBC DB WSDL Web Server WSDL JDBC DB

Solution Part II: Grid Resources n Many Grid tools provide powerful backend services q Globus: uniform, secure access to computing resources (like Tera. Grid) n q q q n File management, resource allocation management, etc. Condor: job scheduling on computer clusters and collections SRB: data grid access OGSA-DAI: uniform Grid interface to databases. These have Web Service as well as other interfaces (or equivalently, protocols).

Solution, Part III: Domain Specific Tools and Standards -->More For Chemical Informatics, we have a number of tools and Services standards. n q Chemical string representations n q Chemistry Markup Language n n q n n XML language for describing, exchanging data. JUMBO 5: a CML parser and library Glue Tools and Applications n n SMILES, In. Ch. I Chemistry Development Kit (CDK) Open. Babel These are the basis for building interoperable Chemical Informatics Web Services Analogous situations exist for other domains q Astronomy, Geosciences, Biology/Bioinformatics

Solution Part IV: Workflows n n n Workflow engines allow you to connect services together into interesting composite applications. This allows you to directly encode your scientific use case scenario as a graph of interacting services. There are many workflow tools q q q We’ll briefly cover these later. General guidance is to build web services first and then use workflow tools on top of these services. Don’t get married to a particular workflow technology yet, unless someone pays you.

Solution Part V: User Interfaces n Web Services allow you to cleanly separate user interfaces from backend services. q n Client environments include q q q n n Model-view-controller pattern for web applications Grid and web service scripting environments Desktop tools like Taverna and Kepler Portlet-based Web portal systems Typically, desktop tools like Taverna are used by power users to define interesting workflows. Portals are for running canned workflows.

Next steps n n Next we will review the online data base resources that are available to us. Databases come in two varieties q q n Journal databases Data databases As we will discuss, it is useful to build services and workflows for automatically interacting with both types.

Online Chemical Journal and Data Resources

MEDLINE: Online Journal Database n n MEDLINE (Medical Literature Analysis and Retrieval System Online) is an international literature database of life sciences and biomedical information. It covers the fields of medicine, nursing, dentistry, veterinary medicine, and health care. MEDLINE covers much of the literature in biology and biochemistry, and fields with no direct medical connection, such as molecular evolution. It is accessed via Pub. Med. http: //en. wikipedia. org/wiki/Medline

n n n Pub. Med: Journal Search Engine Pub. Med is a free search engine offered by the United States National Library of Medicine as part of the Entrez information retrieval system. The Pub. Med service allows searching the MEDLINE database. q MEDLINE covers over 4, 800 journals published in the United States and more than 70 other countries primarily from 1966 to the present. In addition to MEDLINE, Pub. Med also offers access to: q OLDMEDLINE for pre-1966 citations. q Citations to articles that are out-of-scope (e. g. , general science and chemistry) from certain MEDLINE journals q In-process citations which provide a record for an article before it is indexed with Me. SH and added to MEDLINE q Citations that precede the date that a journal was selected for MEDLINE indexing q Some life science journals http: //www. ncbi. nlm. nih. gov/entrez/query/static/overview. html

Pub. Chem: Chemical Database n n n Pub. Chem is a database of chemical molecules. The system is maintained by the National Center for Biotechnology Information (NCBI) which belongs to the United States National Institutes of Health (NIH). Pub. Chem can be accessed for free through a web user interface. q And Web Services for programmatic access n Pub. Chem contains mostly small molecules with a molecular mass below 500. n Anyone can contribute q q The database is free to use, but it is not curated, so value of a specific compound information could be questionable. NIH funded HTS results are (intended to be) available through pubchem. http: //pubchem. ncbi. nlm. nih. gov/

NIH DTP Database n n Part of NIH’s Developmental Therapeutics Program. Screens up to 3, 000 compounds per year for potential anticancer activity. Utilizes 59 different human tumor cell lines, representing leukemia, melanoma and cancers of the lung, colon, brain, ovary, breast, prostate, and kidney. DTP screening results are part of Pub. Chem and also available as a separate database. http: //dtp. nci. nih. gov/

Example screening results. Positive results (red bar to right of vertical line) indicates greater than average toxicity of cell line to tested agent. http: //dtp. nci. nih. gov/docs/compare. html

DTP and COMPARE n n COMPARE is an algorithm for mining DTP result data to find and rank order compounds with similar DTP screening results. Why COMPARE? q q q Discovered compounds may be less toxic to humans but just as effective against cancer cell lines. May be much easier/safer to manufacture. May be a guide to deeper understanding of experiments http: //dtp. nci. nih. gov/docs/compare_methodology. html

Many Other Online Databases n n Complementary protein information Indiana University: Varuna project q n University of Michigan: Binding MOAD q q n Discussed in this presentation “Mother of All Databases” Largest curated database of protein-ligand complexes Subset of protein databank Prof. Heather Carlson University of Michigan: PDBBind q q Provides a collection of experimentally measured binding affinity data (Kd, Ki, and IC 50) exclusively for the proteinligand complexes available in the Protein Data Bank (PDB) Dr. Shaomeng Wang

The Point Is… n All of these databases can be accessed on line with human-usable interfaces. q n More importantly, many of them are beginning to define Web Service interfaces that let other programs interact with them. q n n But that’s not so important for our purposes Plenty of tools and libraries can simulate browsers, so you can also build your own service. This allows us to remotely analyze databases with clustering and other applications without modifying the databases themselves. Can be combined with text mining tools and web robots to find out who else is working in the area.

Encoding chemistry

Chemical Machine Languages n Interestingly, chemistry has defined three simple languages for encoding chemical information. q q n In. Ch. Is and SMILES can represent molecules as a single string/character array. q n Useful as keys for databases and for search queries in Google. You can convert between SMILES and In. Ch. Is q n In. Ch. I, SMILES, CML Can generate these by hand or automatically Open. Babel, OELib, JOELib CML is an XML format, and more verbose, but benefits from XML community tools

SMILES: Simplified Molecular Input Line Entry Specification n Language for describing the structure of chemical molecules using ASCII strings. http: //www. daylight. com/dayhtml/doc/theory. smiles. html

In. C� h. I: International Chemical Identifier n n n IUPAC and NIST Standard similar to SMILES Encodes structural information about compounds Based on open an standard and algorithms. http: //wwmm. ch. cam. ac. uk/inchifaq/

In. Ch. I in Public Chemistry US National Institute of Standards and Technology (NIST) - 150, 000 Databases structures n n n NIH/NCBI/Pub. Chem project - >3. 2 million structures Thomson ISI - 2+ million structures US National Cancer Institute(NCI) Database - 23+ million structures US Environmental Protection Agency(EPA)-DSSTo. X Database - 1450 structures Kyoto Encyclopaedia of Genes and Genomes (KEGG) database - 9584 structures University of California at San Francisco ZINC - >3. 3 million structures BRENDA enzyme information system (University of Cologne) - 36, 000 structures Chemical Entities of Biological Interest (Ch. EBI) database of the European Bioinformatics Institute - 5000 structures University of California Carcinogenic Potency Project - 1447 structures Compendium of Pesticide Common Names - 1437 (2005 -03 -03) structures

Journals and Software Using In. Ch. I n Journals q q n Nature Chemical Biology. Beilstein Journal of Organic Chemistry Software q q ACD/Labs ACD/Chem. Sketch. Chem. Axon Marvin. Sci. Tegic Pipeline Pilot. CACTVS Chemoinformatics Toolkit by Xemistry, Gmb. H. http: //wwmm. ch. cam. ac. uk/inchifaq/

Chemistry Markup Language n CML is an XML markup language for encoding chemical information. q q n More verbose than In. Ch. I and SMILES q n n Developed by Peter Murray Rust, Henry Rzepa and others. Actually dates from the SGML days before XML But inherits XML schema, namespaces, parsers, XPATH, language binding tools like XML Beans, etc. Not limited to structural information Has Open. Babel support. http: //cml. sourceforge. net/, http: //cml. sourceforge. net/wiki/index. php/Main_Page

In. Ch. I Compared to SMILES n n SMILES is proprietary and different algorithms can give different results. Seven different unique SMILES for caffeine on Web sites: q q q q [c]1([n+]([CH 3])[c]([c]2([c]([n+]1[CH 3])[n] [c. H][n+]2[CH 3]))[O-] CN 1 C(=O)N(C)C(=O)C(N(C)C=N 2)=C 1 2 Cn 1 cnc 2 n(C)c(=O)c 12 Cn 1 cnc 2 c 1 c(=O)n(C)c(=O)n 2 C N 1(C)C(=O)N(C)C 2=C(C 1=O)N(C)C=N 2 O=C 1 C 2=C(N=CN 2 C)N(C(=O)N 1 C)C CN 1 C=NC 2=C 1 C(=O)N(C)C(=O)N 2 C On the other hand, some claim SMILES are more intuitive for human readers. http: //wwmm. ch. cam. ac. uk/inchifaq/

A CML Example http: //www. medicalcomputing. net/xml_biosciences. html

Clustering Techniques, Computing Requirements, and Clustering Services Computational techniques for organizing data

The Story So Far n We’ve discussed managing screening assay output as the key problem we face q q n Must sift through mountains of data in Pub. Chem and DTP to find interesting compounds. NIH funded High Throughput Screening will make this very important in the near future. Need now a way to organize and analyze the data.

Clustering and Data Analysis n n n Clustering is a technique that can be applied to large data sets to find similarities q Popular technique in chemical informatics Data sets are segmented into groups (clusters) in which members of the same cluster are similar to each other. Clustering is distinct from classification, q There are no pre-determined characteristics used to define the membership of a cluster, q Although items in the same cluster are likely to have many characteristics in common. Clustering can be applied to chemical structures, for example, in the screening of combinatorial or Markush compound libraries in the quest for new active pharmaceuticals. We also note that these techniques are fairly primitive q More interesting clustering techniques exist but apparently are not well known by the chemical informatics community.

Non-Hierarchical Clustering n n n Clusters form around centroids. The number of which can be specified by the user. All clusters rank equally and there is no particular relationship between them. http: //www. digitalchemistry. co. uk/prod_clustering. html

Hierarchical Clusters are arranged. Clustering in hierarchies n q q Smaller clusters are contained within larger ones; the bottom of the hierarchy consists of individual objects in "singleton" clusters, while the top of it consists of one cluster containing all the objects in the dataset. Such hierarchies can be built either from the bottom up (agglomerative) or the top downwards (divisive) http: //www. digitalchemistry. co. uk/prod_clustering. html

Fingerprinting and Dictionaries--What Is Your Parameter Space? n Clustering algorithms require a parameter space q n n Clusters defined along coordinate axes. Coordinate axes defined by a dictionary of chemical structures. Use binary on/off for fingerprinting a particular compound against a dictionary. http: //www. digitalchemistry. co. uk/prod_fingerprint. html

Cluster Analysis and Chemical Informatics n n Used for organizing datasets into chemical series, to build predictive models, or to select representative compounds Clustering Methods q Jarvis-Patrick and variants n q Ward’s method n q < O(N 2), requires set no of clusters, a little “messy” Sphere-exclusion (Butina) n q Hierarchical, regarded as best, but at least O(N 2) K-means n q O(N 2), single partition Fast, simple, similar to JP Kohonen network n Clusters arranged in 2 D grid, ideal for visualization

Limitations of Ward’s method for large datasets (>1 m) n n Best algorithms have O(N 2) time requirement (RNN) Requires random access to fingerprints q n Problem of selection of best partition q n hence substantial memory requirements (O(N)) can select desired number of clusters Easily hit 4 GB memory addressing limit on 32 bit machines q Approximately 2 m compounds

Scaling up clustering methods n Parallelization q q q n Clustering algorithms can be adapted for multiple processors Some algorithms more appropriate than others for particular architectures Ward’s has been parallelized for shared memory machines, but overhead considerable New methods and algorithms q q q Divisive (“bisecting”) K-means method Hierarchical Divisive Approx. O(Nlog. N)

Divisive K-means Clustering n New hierarchical divisive method q q q n Hierarchy built from top down, instead of bottom up Divide complete dataset into two clusters Continue dividing until all items are singletons Each binary division done using K-means method Originally proposed for document clustering “Bisecting K-means” q q q Steinbach, Karypis and Kumar (Univ. Minnesota) http: //wwwusers. cs. umn. edu/~karypis/publications/Papers/PDF/docclu ster. pdf Found to be more effective than agglomerative methods Forms more uniformly-sized clusters at given level

BCI Divkmeans n Several options for detailed operation q q q n Options within each K-means division step q q n n Selection of next cluster for division size, variance, diameter affects selection of partitions from hierarchy, not shape of hierarchy distance measure choice of seeds batch-mode or continuous update of centroids termination criterion Have developed parallel version for Linux clusters / grids in conjunction with BCI For more information, see Barnard and Engels talks at: http: //cisrg. shef. ac. uk/shef 2004/conference. htm

Comparative execution times NCI subsets, 2. 2 GHz Intel Celeron processor 7 h 27 m 3 h 06 m 2 h 25 m 44 m

Divisive K-means: Conclusions n Much faster than Ward’s, speed comparable to K-means, suitable for very large datasets (millions) q q q n n n Time requirements approximately O(N log N) Current implementation can cluster 1 m compounds in under a week on a low-power desktop PC Cluster 1 m compounds in a few hours with a 4 -node parallel Linux cluster Better balance of cluster sizes than Wards or Kmeans Visual inspection of clusters suggests better assembly of compound series than other methods Better clustering of actives together than previouslystudied methods Memory requirements minimal Experiments using AVIDD cluster and Teragrid forthcoming (50+ nodes)

Conclusions n n n Effective exploitation of large volumes and diverse sources of chemical information is a critical problem to solve, with a potential huge impact on the drug discovery process Most information needs of chemists and drug discovery scientists are conceptually straightforward, but complex to implement All of the technology is now in place to implement may of these information need “use-cases”: the four level model using service -oriented architectures together with smart clients look like a neat way of doing this In conjunction with grid computing, rapid and effective organization and visualization of large chemical datasets is feasible in a web service environment Some pieces are missing: q q Chemical structure search of journals (wait for In. Ch. I) Automated patent searching Effective dataset organization Effective interfaces, especially visualization of large numbers of 2 D structures

Divisive K-Means as a Web Service n The previous exercise was intended to show that Divisive K-Means is a classic example of Grid application. q q n n Needs to be parallelized Should run on Tera. Grid How do you make this into a service? We’ll go on a small tour before getting back to our problem.

Wrapping Science Applications as Services n n Science Grid services typically must wrap legacy applications written in C or Fortran. You must handle such problems as q Specifying several input and output files n q q n n These may need to be staged in Launching executables and monitoring their progress. Specifying environment variables Often these have also shell scripts to do some miscellaneous tasks. How do you convert this to WSDL? q Or (equivalently) how do you automatically generate the XML job description for WS-GRAM?

Generic Service Toolkit (GFAC) (G. Kandaswamy, IU and RENCI) n n The Generic Service Toolkit can "wrap" any command-line application as an application service. q Given a set of input parameters, it runs the application, monitors the application and returns the results. Requires no modification to program code. Also has web user interface generating tools. q When a user accesses an application service, the user is presented with a graphical user interface (GUI) to that service. q The GUI contains a list of operations that the user is allowed to invoke on that service. After choosing an operation, the user is presented with a GUI for that operation, which allows the user to specify all the input parameters to that operation. q The user can then invoke the operation on the service and get the output results. www. extreme. indiana. edu/gfac/

OPAL (S. Krishan, SDSC) n n Features include scheduling (using Globus and Condor/SGE) and security (using GSI-based certificates), and persistent state management. The WSDL defines operations to do the following: q q q q get. App. Metadata: includes usage information, arbitrary application-specific metadata specified as an array of other elements, n e. g. description of the various options that are passed to the application binary. launch. Job: runs job with specified input and returns a Job ID. query. Status: returns status code, message, and URL of the working directory get. Outputs: returns the outputs from a job that is identified by a Job ID. n URLs for the standard output and error n Array of structures representing the output file names and URLs get. Output. As. Base 64 By. Name: This operation returns the contents of an output file as Base 64 binary. destroy: This operation destroys a running job identified by a Job ID. launch. Job. Blocking: This operation requires the list of arguments as a string, and an array of structures representing the input files. http: //grid-devel. sdsc. edu/gridsphere? cid=nbcrws

Our Solution: Apache Ant Services n We’ve found using Apache Ant to be very useful for wrapping services. q q n Can call executables, set environment variables. Lots of useful built-in shell-like tasks. Extensible (write your own tasks). Develop build scripts to run your application You can easily call Ant from other Java programs. q q q So just write a wrapper service We use both blocking (hold connection until return) and non -blocking version (suitable for long running codes). In non-blocking case, “Context” web service is used for callbacks.

Flow Chart of SMILES to Cluster Partitioned of BCI Web Service SMILES to DKM SMILE String Makebits Generating Fingerprints Dictionary (Default) Fingerprint (*. scn) Generating the best levels Div. Kmeans Clustering Fingerprints Extracting individual cluster partitions Optclus best level RNNclus Extracted Cluster Hierarchy (*. clu) One Column Process Cluster Hierarchy (*. dkm) New SMILE String Merge Process

BCI Clustering Service Methods Service Method Description Input Output makebits. Generate fingerprints SMIstring from a SMILES structure Fingerprint string divkm. Generate Cluster fingerprints with Divkmeans SCNstring Clustered Hierarchy smile 2 dkm Makebits + divkm SMIstring optclus. Generate the best levels DKMstring Best partition in a hierarchy cluster level rnnclus. Generate Extract individual cluster partitions Clustered Hierarchy DKMstring Indiv. cluster partitions smile 2 Cluster. Partiti Generate a new SMILES SMIstring oned structure w/ extra col. New SMILES structure

A Library of Chemical Informatics Web Services

All Services Great and Small n Like most Grids, a Chemical Informatics Grid will have the classic styles: q q n But we also need many additional services q q n Data Grid Services: these provide access to data sources like Pub. Chem, etc. Execution Grid Services: used for running cluster analysis programs, molecular modeling codes, etc, on Tera. Grid and similar places. Handling format conversions (In. Ch. I<->SMILES) Shipping and manipulating tabular data Determining toxicity of compounds Generating batch 2 D images So one of our core activities is “build lots of services”

VOTables: Handling Tabular Data n n n Developed by the Virtual Observatory community for encoding astronomy data. The VOTable format is an XML representation of the tabular data (data coming from BCI, NIH DTP databases, and so on). VOTables-compatible tools have been built q n We just inherit them. SAVOT and JAVOT JAVA Parser APIs for VOTable allow us to easily build VOTable-based applications q q q Web Services Spread sheet Plotting applications. n VOPlot and Top. Cat are two

Document Structure of VOTable Compound Name Cluster Number Acemetacin 1 Candesartan 1 Acenocoumarol 2 Dicumarol 2 Phenprocoumon 2 Trioxsalen 2 Warfarin 2 <? xml version="1. 0"? > <VOTABLE version="1. 1“ xmlns: xsi=http: //www. w 3. org/2001/XMLSchemainstance xsi: no. Namespace. Schema. Location="http: //www. ivoa. net/xml/VOTable/VOTa ble/v 1. 1"> <RESOURCE > <TABLE name="results"> <FIELD name=“Compound. Name" ID="col 1" datatype=“char" arraysize=“*”/> <FIELD name=“Clusture. Number” ID="col 2“ datatype=“int”/> <DATA> <TABLEDATA> <TR><TD>Acemetacin</TD><TD>1</TD</TR> <TR><TD>Candesartan</TD><TD>1</TD></TR> <TR><TD>Acenocoumarol</TD><TD>2</TD></TR> <TR><TD>Dicumarol</TD><TD>2</TD></TR> <TR><TD>Phenprocoumon</TD><TD>2</TD></TR> <TR><TD>Trioxsaken</TD><TD>2</TD></TR> <TR><TD>warfarin</TD><TD>2</TD></TR> </TABLEDATA> </TABLE> </RESOURCE> </VOTABLE>

mrtd 1. txt – smiles representation of chemical compounds along with its properties

mrtd 1. txt Taverna Client WSDL VOTable. Generator. Service Tomcat Server VOTable. Generator. Servic e retrieve. VOTable. Document votable. xml VOPlot

Votable. xml : xml representation of mrtd 1. txt file

VOPlot Application from generated votable. xml file : Graph plotted on Mass (X–axis) and PSA (Y-axis)

Other Uses for VOTables n n VOTables is a useful intermediate format for exchanging data between data bases. Simple example: exchange data between VARUNA databases. q q n Each student in the Baik group maintains his/her on copy (sandbox purposes). Often need to import/export individual data sets. It is also good for storing intermediate results in workflows. q Value is not the format, but the fact that the XML can be manipulated programmatically. n Unions, subset, intersection operations

More Services: WWMM Services Descriptions Input Output In. Ch. IGoogle Search an In. Ch. I inchi. Basic structure through Google type Search result in HTML format In. Ch. IServer version format An In. Ch. I structure Open. Babel. S Transform a chemical erver format to another using Open Babel format input. Data output. Data options Converted chemical structure string CMLRSSSer Generate CMLRSS feed ver from CML data mol, title Converted description CMLRSS feed link, source of CML data Generate In. Ch. I

CDK-Based Services Common Substructure Calculates the common substructure between two molecules. CDKsim Takes two SMILES and evaluates the Tanimoto coefficient (ratio of intersection to union of their fingerprints). CDKdesc Calculates a variety of molecular and atomic descriptors for QSAR modeling CDKws Fingerprint generation CDKsdg Creates a jpeg of the compound’s 2 D structure CDKStruct 3 D Generates 3 D coordinates of a molecule from its SMILE

Tox. Tree Service n n n The Threshold of Toxicological Concern (TTC) establishes a level of exposure for all chemicals below which there would be no appreciable risk to human health. Tox. Tree implements the Cramer Decision Tree approach to estimate TTC. We have converted this into a service. q q Uses SMILES as input. Note the GUI must be separated from the library to be a service http: //ecb. jrc. it/QSAR/home. php? CONTENU=/QSAR/qsar_tools_toxtree. php

tox. Tree

Taverna Workflow for Toxic Hazard Estimation

OSCAR 3 Service n n Oscar 3 is a tool for shallow, chemistry-specific natural language parsing of chemical documents (i. e. journal articles). It identifies (or attempts to identify): q q q n There is a larger effort, Sci. Borg, in this area q n n Chemical names: singular nouns, plurals, verbs etc. , also formulae and acronyms. Chemical data: Spectra, melting/boiling point, yield etc. in experimental sections. Other entities: Things like N(5)-C(3) and so on. http: //www. cl. cam. ac. uk/~aac 10/escience/sciborg. html This (like Tox. Tree) is potentially productively pleasingly parallelized. It also has potentially very interesting Workflows http: //wwmm. ch. cam. ac. uk/wikis/wwmm/index. php/Oscar 3

Use Cases and Workflows Putting data and clustering together in a distributed environment.

Chemical Informatics as a Grid Problem NIH-Funded experimental screening n q q n n n NIH DTP and HTS projects are generating a wealth of raw data on small compounds. Available in Pub. Chem Journal and chemical data sources often have public Web clients and GUIs. q But we need Web Service interfaces, not just Web interfaces. q These provide a programming interfaces for building both human and machine clients. These need to be connected to computing resources for running clustering, data mining, and molecular modeling applications. q Excellent candidates for running on the Tera. Grid We can formulate scientific problems that map to interconnections of Grid services. q This is generally called “Grid workflow” or “Service Orchestration”

Computation Oracle Database (HTS) Compounds were tested against related assays and showed activity, including selectivity within target families All the compounds pass the Lipinksi Rule of Five and toxicity filters Excel Spreadsheet (Toxicity) One of the compounds was previously tested for toxicology and was found to have no liver toxicity Oracle Database (Genomics) ? None of these compounds have been tested in a microarray assay ? Computation Several of the compounds had been followed up in a previous project, and solubility problems prevented further development Journal Article SCIENTIST The information in the structures and known activity data is good enough to create a QSAR model with a confidence of 75% Word Document (Chemistry) A recent journal article “These compounds look promising from their HTS results. Should I commit some chemistry resources to following them up? ” reported the effectiveness of some compounds in a related series against a target in the same family External Database (Patent) Word Document (Marketing) Some structures with a A report by a team in Marketing casts doubt on whether the market for this target is big enough to make development cost-effective similarity > 0. 75 to these appear to be covered by a patent held by a competitor

Workflow, Services, and Science n Web Services work best as simple stateless services. q n n n No implicit input, output, or interdependency of methods. Services must be composed into interesting applications. This is called workflow. A good workflow. . . q q Is composed of independent services Completely specifies an interesting science problem.

n Some Open Source Grid Workflow Projects UK e-Science Project’s Taverna q n n n Scufl. xml scripting, GUI interface, works with Web Services. Kepler q Works with Web services and the Globus Toolkit. Condor DAGMan q www. cs. wisc. edu/condor q Works over the top of Condor’s scheduler. q Extended by the Gri. Phy. N Virtual Data System Java Co. GKit’s Karajan q XML workflow specification for scripting COG clients. q Works with GT 2 and 4. Community Grids Lab’s HPSearch q www. hpsearch. org q Java. Script scripting, works with Web services. Indiana Extreme Lab’s Workflow Composer q www. extreme. indiana. edu/xgws/xwf/index. html q Jython, BPEL (soon) scripting

Finding compound-protein relationships A 2 D structure is supplied for input into the similarity search (in this case, the extracted bound ligand from the PDB IY 4 complex) A protein implicated in tumor growth is supplied to the docking program (in this case HSP 90 taken from the PDB 1 Y 4 complex) The workflow employs our local NIH DTP database service to search 200, 000 compounds tested in human tumor cellular assays for similar structures to the ligand. Client portlets are used to browse these structures Similar structures are filtered for drugability, and are automatically passed to the Open. Eye FRED docking program for docking into the target protein. Correlation of docking results and “biological fingerprints” across the human tumor cell lines can help identify potential mechanisms of action of DTP compounds Once docking is complete, the user visualizes the highscoring docked structures in a portlet using the JMOL applet.

HTS data organization & flagging Open. Eye FILTER is used to calculate biological and chemical properties of the compounds that are related to their potential effectiveness as drugs The compounds are clustered on chemical structure similarity, to group similar compounds together A tumor cell line is selected. The activity results for all the compounds in the DTP database in the given range are extracted from the Postgre. SQL database The compounds along with property and cluster information are converted to VOTABLES format and displayed in VOPLOT

Use Case: Which of these hits should I follow up? n An HTS experiment has produced 10, 000 possible hits out of a screening set of 2 m compounds. A chemist on the project wants to know what the most promising series of compounds for follow-up are, based on: q q q q Series selection cluster analysis Structure-activity relationships modal fingerprints/stigmata Chemical and pharmacokinetic properties mitools, chemaxon Compound history g. Nova / Postgre. SQL Patentability BCI Markush handling software Toxicity Synthetic feasibility + requires visualization tools!

A Workflow Scenario: HTS Data Organization and Flagging n n n This workflow demonstrates how screening data can be flagged and organized for human analysis. The compounds and data values for a particular screen are retrieved from the NIH DTP database and then are filtered to remove compounds with reactive groups, etc. q A tumor cell line is selected. The activity results for all the compounds in the DTP database in the given range are extracted from the Postgre. SQL database Open. Eye FILTER is used to calculate biological and chemical properties of the compounds that are related to their potential effectiveness as drugs Tox. Tree is used to flag the potential toxicities of compounds. Divkmeans is used to add a column of cluster numbers. Finally, the results are visualized using VOPlot and the 2 D viewer applet.

Web Services

Example plots of our workflow output using VOPlot and VOTables

SMILES + ID + Data NIH Database Service SMILES + ID Fingerprint Generator Postgre. SQL CHORD BCI Makebits Cluster Analysis Fingerprints Me Clus mb ter ers hip BCI Divkmeans Table Management Vo. Tables Cluster the compounds in the NIH DTP database by chemical structure, then choose representative compounds from the clusters and dock them into PDB protein files of interest SMILES + ID + + Cluster # + Data Docking Selector 3 D Visualizer JMOL Script Plot Visualizer Vo. Plot SMILES + ID Do Co cked mp lex Docking 2 D-3 D Open. Eye FRED MOL File Open. Eye OMEGA PDB Database Service PDB Structure + Box

Use Case: Are there any good ligands for my target? n A chemist is working on a project involving a particular protein target, and wants to know: q q Any newly published compounds which might fit the protein receptor site g. Nova / Postgre. SQL, Pub. Chem search, FRED Docking Any published 3 D structures of the protein or of proteinligand complexes PDB search Any interactions of compounds with other proteins g. Nova / Postgre. SQL, Pub. Chem search Any information published on the protein target Journal text search

Use Case: Who else is working on these structures? n A chemist is working on a chemical series for a particular project and wants to know: q q q If anyone publishes anything using the same or related compounds ~ Pub. Chem search Any new compounds added to the corporate collection which are similar or related g. Nova CHORD / Postgre. SQL If any patents are submitted that might overlap the compounds he is working on ~ BCI Markush handling software Any pharmacological or toxicological results for those or related compounds g. Nova CHORD / Postgre. SQL, Mi. Toolkit The results for any other projects for which those compounds were screened g. Nova CHORD / Postgre. SQL, Pub. Chem search

VARUNA – Towards a Gridbased Molecular Modeling Environment A brief overview of Prof. Mookie Baik’s VARUNA project.

Chemical Informatics in Academic Research? n Industrial Research: Target n Academic Research: Oriented q q q q Not bound to a specific molecular system Not bound to a method Not concerned with generality Aware of Efficiency Aware of Overall Cost Aware of Toxicity Concerned about Formulations Cares about active MOLECULES Concept Oriented q q q q Specialized on few molecular families Method Development is important Obsessed with generality Does not care much about efficiency Cost is unimportant Often can’t even assess for Toxicity Formulation is a minor issue Cares mostly about REACTIONS, i. e. ways to GET to a molecule

Auto. Ge. FF, Varuna and Workflows n Metalloproteins are extremely important in n biochemical processes Understanding their chemistry is difficult To add value to the small molecule DB’s (Pub. Chem, etc. ), we must somehow connect them to PDB’s, Bind. MOAD, etc. By extending Varuna’s functionality to handling, storing Metalloproteins, we could provide a connection

Automatic Generator of Force. Fields (Auto. Ge. FF) n n Developing a service that can take ANY q drug-like molecule (from Pub. Chem, for example) q metal complexes q metalloenzymes (from PDB, for example) q unnatural or functionalized amino acids, nucleobases (from inhouse db) for which molecular mechanics force fields are not available and automatically generate FF’s based on q High level Quantum Simulations (using Varuna as a Web service) for Sophisticated Molecular Mechanics Simulations First Step: Coding of a specialized Prototype that can reproduce our manually derived novel force fields for Cu-Ab Alzheimer’s Disease as a Proof-Of-Principles Study.

Automatic Quantum Mechanical Curation of Structure Data n n n Chemical Research logic is often driven by molecular structure Large-scale, small molecule DB’s (such as Pub. Chem) have low-resolution structure data Often key properties are not consistently available: q n e. g. : Rotation-barriers, Redox Potentials, Polarizabilities, IR frequencies, reactivity towards nucleophiles QM web-services will provide tools for generating highresolution data q q q that will curate the results of traditional Chem. Info studies allow for combinatorial computational chemistry access a database of modeling data

Prototype-Project: Controlling the TGFb pathway in-house Molecules in Varuna Simulations Auto. Ge. FF VARUNA Conceptual Understanding of TGFb Inhibition Inactive TGFb 1 IAS Active TGFb With inhibitor Questions: - What molecular feature controls inhibitor binding? Pub. Chem PDB Experiments in the Zhang Lab - How do mutations impact binding?

Consequences for Chem. Info Design for Academia n n TWO Strategies are needed: q Making traditional Chem. Info tools that are often available in commercial research available to Academia is in principle straightforward. q New Chem. Info Tools that are CONCEPT centered and include REACTIONS in addition to MOLECULES must be developed. Our approach: Development of (a) Quantum Chemical Database (b) Molecular Modeling Database q Harness the power of recent advances in Molecular Modeling (QM, QM/MM, MD) through information management. q Data-depository for Quantum Chemical Data including both Properties & Mechanisms

QM Calculation Workflow

More Information n n Contact me: mpierce@cs. indiana. edu Most of this was taken from our CICC project. See www. chembiogrid. org/wiki. q q n Many elements of our approach are based on Prof. Peter Murray Rust’s group’s approach. q n Note we’ve found wikis to be extremely useful and fun to use for maintaining collaborative web sites. See also www. crisisgrid. org and www. gorerle. com/vlab-wiki for other examples using Media Wiki. WWMM Wiki: wwmm. ch. cam. ac. uk/wikis/wwmm/index. php Source. Forge Project Site q http: //sourceforge. net/projects/cicc-grid

Additional Slides

Use Case - CICC Which of these hits should I follow An MLI HTS experiment has produced 10, 000 possible hits out of up? a screening set of 2 m compounds. A chemist at another n laboratory wants to know if there any interesting active series she might want to pursue, based on: q Structure-activity relationships q Chemical and pharmacokinetic properties q Compound history q Patentability q Toxicity q Synthetic feasibility

CICC Web Services I n BCI Clustering q q q Provides Bernard Chemical Information (BCI) clustering packages A module of the workflow for HTS data organization and flagging Status: n n q n Added URL output support to the previous solid prototype (Multi-user durable) Taverna Beanshell Scripting for data format adjusting (e. g. Filtering out the head part listing column names) To do: Evaluating the URI(URL) based workflow design Tox. Tree q q Estimates toxic hazard by applying a decision tree approach A module of the workflow for HTS data organization and flagging Status: A test prototype producing the level of toxicity in a brief or verbose explanation against a SMILE structure To do: n n Refining the Web service for cluster input and external property support The Taverna Beanshell scripting for data merging not used in some modules

CICC Web Services II n Workflow for HTS data organization and flagging q q q Demonstrates how screening data can be flagged and organized for human analysis Status: Individual modules except the visualization are in prototype To do: n n n Defining at least XML schema or DTD for the workflow data (at most the Ontology) Redefining current workflow model to reflect the new feature of Taverna 1. 4 supporting complex data structures and the provenance plugin Other Planed Web Services q Open Source Chemistry Analysis Routines (OSCAR) n n q Extracts chemical information from text and produces an XML instance highlighting the chemical information A module of the PMR workflow Status: OSCAR 3 is available and works fine as a Java application To do: Studying XML instances for extracting chemical names Info. Chem’s SPRESI Web Service n n n Provides access to the SPRESI molecule database Status: Perl scripts for accessing SPRESI Web Service To do: Developing a Web service wrapper to utilize Info. Chem’s SPRESI Web Service

BCI Clustering URL Service Methods Service Method Description Input makebits. URLGenerate fingerprints SMIstring rate from a SMILES structure URLOutput Fingerprint and program output divkm. URLGenerat e Cluster fingerprints with Divkmeans SCNstring DKM data and program output smile 2 dkm. URL Makebits + divkm SMIstring All SMI, DKM and std. outputs optclus. URLGenerate the best levels SMIstring Best data and te in a hierarchy DKMstring program output rnnclus. URLGenera Extract individual cluster te partitions SMIstring New partition DKMstring and std. output smile 2 Cluster. Partiti Generate a new SMILES SMIstring oned. URL structure w/ extra col. All intermediate data and output

Workflow for smile 2 Cluster. Partitioned. URL

Workflow for Toxic Hazard in Verbose

Diagram of Workflow 2 Input/Output Web Services Beanshell Scripting

Informatics is the discipline of science which investigates the structure and properties (not specific content) of scientific information, as well as the regularities of scientific information activity, its theory, history, methodology and organization. The purpose of informatics consists in developing optimal methods and means of presentation (recording), collection, analyticalsynthetic processing, storage, retrieval and dissemination of scientific information. A. I. Mikhailov, A. I. Chernyi, R. S. Gilyarevskii (1967) “Informatics -New Name of the Theory of Scientific Information”

Chemical informatics is … n n More usually know as chemoinformatics or cheminformatics Very differently defined, reflecting its crossdisciplinary nature q q q Librarian Chemist (synthetic, medicinal, theoretical) Biologist / Bioinformatician Molecular modeler Pharmaceutical or Chemical Engineer Computer Scientist / Informatician

More definitions n n n Computational Chemistry – The application of mathematical and computational methods to particularly to theoretical chemistry Molecular Modeling – Using 3 D graphics and optimization techniques to help understand the nature and action of compounds and proteins Computer-Aided Drug Design – The discipline of using computational techniques (including chemical informatics) to assist in the discovery and design of drugs.

Traditional areas of application n Pharmaceutical & life science industry q n n Databases of available chemicals Electronic publishing q n particularly in early stage drug design including searchable chemical structure information in journals, etc. Government and patent databases

The –ics so far (1960’s to present) … n How do you represent 2 D and 3 D chemical structures? q n How do you search databases of chemical structures? q n n n Not just a pretty picture Google doesn’t help (much, but it might do soon…) How do you organize large amounts of chemical information? How do you visualize chemical structures & proteins? Can computers predict how chemicals are going to behave q q … in the test tube? … in the body?

Current trends & hot topics n n The decorporatization of chemical informatics (Pub. Chem, MLI, e. Science, open source) Service-oriented architectures Packaging & processing large volumes of complex information for human consumption Integration with other –ics (bioinformatics, genomics, proteomics, systems biology)

Main players (Commercial) MDL www. mdl. com n Tripos, inc. www. tripos. com n Accelrys www. accelrys. com n Daylight CIS, inc. www. daylight. com n

Main players (Academia) n “Pure” Chemoinformatics q University of Sheffield, UK (Willett / Gillet) n q Erlangen, Germany (Gasteiger) n q http: //www-ucc. ch. cam. ac. uk/ Indiana University School of Informatics n n http: //www 2. chemie. uni-erlangen. de/ Cambridge Unilever Center n q http: //www. shef. ac. uk/uni/academic/I-M/is/research/cirg. html http: //www. informatics. indiana. edu/ Related (computational chemistry, etc. ) q UCSF (Kuntz) n q University of Texas (Pearlman) n q http: //www. utexas. edu/pharmacy/divisions/pharmaceutics/faculty/pearlman. html Yale (Jorgensen) n q http: //mdi. ucsf. edu/ http: //zarbi. chem. yale. edu/ University of Michigan (Crippen) n http: //www. umich. edu/~pharmacy/Med. Chem/faculty/crippen/

“Traditional” Journals n Journal of Chemical Information & Modeling (formerly JCICS) q n Journal of Computer-Aided Molecular Design q n http: //www 3. interscience. wiley. com/cgi-bin/jhome/33822 Journal of Chemical Theory and Computation q n http: //www. elsevier. com/inca/publications/store/5/2/5/0/1/2/ Journal of Computational Chemistry q n http: //www. kluweronline. com/issn/0920 -654 X Journal of Molecular Graphics and Modeling q n http: //pubs. acs. org/journals/jcisd 8/index. html http: //pubs. acs. org/journals/jctcce/ Journal of Medicinal Chemistry q http: //pubs. acs. org/journals/jmcmar/

“Informal” publications n Network Science (online) q n Chemical & Engineering News q n http: //www. drugdiscoverytoday. com/ Scientific Computing World q n http: //pubs. acs. org/cen/ Drug Discovery Today q n http: //www. netsci. org/Science/index. html http: //www. scientific-computing. com/ Bio-IT World q http: //www. bio-itworld. com/

CINF-L Distribution List n n n Chemical Information Sources Discussion List Created by Gary Wiggins at IUB http: //www. indiana. edu/~cheminfo/network. ht ml

Yahoo! Chemoinformatics Discussion List n For q q n n Job postings Ideas exchange Questions Industry – Student connections All students encouraged to join Open to others To join, go to http: //groups. yahoo. com/group/chemoinf Or send an email to chemoinf-subscribe@yahoogroups. com

Open Source / Free Software n n n n Blue Obelisk - http: //wiki. cubic. uni-koeln. de/dokuwiki/doku. php In. Ch. I - http: //www. iupac. org/inchi/ JMOL – http: //jmol. sourceforge. net FROWNS - http: //frowns. sourceforge. net/ Open. Babel - http: //openbabel. sourceforge. net/ CML - http: //cml. sourceforge. net/ CDK - http: //almost. cubic. uni-koeln. de/cdk/ MMTK - http: //starship. python. net/crew/hinsen/MMTK/

Example 2 3 D Visualization & Docking 3 D Visualization of interactions between compounds and proteins “Docking” compounds into proteins computationally

3 D Visualization n n X-ray crystallography and NMR Spectroscopy can reveal 3 D structure of protein and bound compounds Visualization of these “complexes” of proteins and potential drugs can help scientists understand the mechanism of action of the drug and to improve the design of a drug Visualization uses computational “ball and stick” model of atoms and bonds, as well as surfaces Stereoscopic visualization available

Docking algorithms n n n Require 3 D atomic structure for protein, and 3 D structure for compound (“ligand”) May require initial rough positioning for the ligand Will use an optimization method to try and find the best rotation and translation of the ligand in the protein, for optimal binding affinity

Genetic Algorithms n n Create a “population” of possible solutions, encoded as “chromosomes” Use “fitness function” to score solutions Good solutions are combined together (“crossover”) and altered (“mutation”) to provide new solutions The process repeats until the population “converges” on a solution

Traditional Workflow of Molecular Modeling FORTRAN Code, Scripts, Visualization Code Researcher Chemical Concepts Experiments Supercomputer Hard Drive Directory Jungle n Highly inefficient workflow (no automation) n Knowledge is human bound (grad student leaves and projects dies) n Incorporation with other DB’s is done in Researcher’s head

Varuna – a new environment for molecular modeling Chemical Concepts Researcher Chem-Grid Experiments Reaction DB QM Database Pub. Chem, PDB, NCI, etc. Simulation Service FORTRAN Code, Scripts DB Service Queries, Clustering, Curation, etc. QM/MM Database Supercomputer

Tools for mining the data Tripos Benchware HTS Dataminer (formerly SAR Navigator), www. tripos. com