BOSC 2006/Abstracts

BioPostgres

D.S. Parker, Ph.D. UCLA

Friday August 4th, 9:00am

BioPostgres is a collection of modules that can be used to extend PostgreSQL, the popular open-source database system, for bioinformatics and computational biology. Each of the BioPostgres modules is intended to be independent, and to be usable separately or in conjunction with the others. Current modules include:

BLASTgres -- Biosequence database extensions GObase -- GeneOntology (GO) extensions PostGraph -- Graph database extensions PostMake -- Data derivation dependency extensions PostModel -- Model base/data mining extensions

Each extension implements new datatypes with query operators and other tools, providing powerful large-scale query and analysis, quick inclusion of new types of biological information, and integration of diverse existing bioscience data. For example, two important datatypes implemented in the modules above are biosequence locations and graphs, and query operators include BLAST access and graph component extraction.

BioPostgres illustrates how the extensibility of PostgreSQL and open-source development can be harnessed for bioscience data management. Modules are added to a user's PostgreSQL source tree as needed; each module is compiled separately and installed in a loadable library directory. PostgreSQL itself is _not_ recompiled, but instead permits library modules to be dynamically linked into a database -- added "on the fly" -- with appropriate SQL commands.

BioPostgres is being developed at the Center for Computational Biology (CCB) at UCLA. Currently all modules are distributed under the GNU GPL.

Web site: www.biopostgres.org

GObase

Ruey-Lung Hsiao, UCLA

Friday August 4th, 9:30am

The Gene Ontology (GO) defines standard vocabulary for biological terms in three aspects -- molecular functions, biological processes, and cellular components. Its wide acceptance and semantic annotation has led to a wide range of applications -- semantic integration, functional analysis, and microarray gene clustering, to name a few. It represents an important new route for connecting information of different types and has become an essential component in system biology.

The success of the GO underscores the importance of having ways to manage, query and visualize it. Despite this importance, most researchers still use an inefficient way to represent and query the GO. The central data structure of the standard relational database for the GO consists of two tables: term and term2term. The term table keeps track of the basic information, such as accession number and term type, of a particular GO term and term2term represents the relationship between two terms (i.e., edges in the term graph). However, most queries for ontology are recursive in nature (One example: find all the descendants of a particular node), this representation in conjunction with SQL can fare poorly under a variety of performance measures.

In order to resolve this issue, the GO database pre-computes the transitive closure of the graph to keep track of every descendants of the node in the graph. In this way, graph traversal queries can be answered through this table without issuing recursive SQL commands. However, the pre-computation of transitive closure leads to very inefficient usage of storage space by storing a huge number of node pairs. There is a clear and significant need for better support of GO representation.

To address these issues we developed the GObase system, a publicly-available open-source platform for the management, query, and visualization of GO information. GObase is a graph database, implemented as an extension of the PostgreSQL database with a graph datatype. This datatype permits storage and (with addition of new functions) sophisticated query of graph data. The graph representation is both more natural and more efficient for queries than the widely-used GO relational database. There are existing browsers for visualizing GO information, but the GO Term Viewer of GObase provides a more powerful interface, allowing both interactive query and interactive annotation of GO terms. In addition, GObase also links the GO with various other biological information resources.

website: www.go-base.org license: GNU General Public License

BioMoby

Martin Senger, Ph.D, International Rice Research Institute, Philippines

Friday August 4th, 10:00am

The Generation Challenge Programme (GCP) is an international agricultural research consortium, currently numbering 20 agricultural research institutes, working on the characterization of plant genetic resources and the application of comparative genomics, toward crop improvement for the developing world. Given the global dispersion of GCP partners, distributed access to informatics resources is a major challenge. Open source projects, such as BioMoby, are essential for linking GCP components into a coherent information gateway.

The BioMoby is an integration tool helping to connect databases, analysis programs, and other resources into a unified distributed system, linking gene discovery with genetic resource characterization and crop evaluation data. GCP developers added new components there, some of them are presented here.

BioMoby Dashboard is a rich standalone Java application for BioMoby developers, assisting through the whole process of creating, deploying and using Biomoby services. It has a plug-in architecture allowing additional extensions.

BioMoby MoSeS ("Moby Service Support") is a set of code generators that transform information and ontology trees stored in a central service registry into Java source code, helping developers of new services to concentrate only on business logic and not on the protocol and messaging details. Such approach guarantees scalability of the developing process.

BioMoby Environment is an automated way to regularly check if the deployed services are running and producing correct results. In the environment with so many participants, such quality control tool is essential.

BioMoby can cooperate with other integrating tools. A separate project "BioCASE & BioMoby" shows how to use BioMoby to "wrap a wrapper". The BioCASE integrates access to many databases, and BioMoby spreads its data into existing clients and networks.

Links: GCP http://www.generationcp.org BioMoby http://biomoby.org BioMoby Java projects: http://biomoby.open-bio.org/CVS_CONTENT/moby-live/Java/docs/ BioCASE http://www.biocase.org

XRATE

Ian Holmes, Ph.D, Berkeley University

Friday August 4th, 11:00am

A big hit in the past couple of years has been the "phylo-HMM", a multi-sequence HMM employing Felsenstein's pruning algorithm to compute emission scores. (Stochastic grammar extensions to this idea include phylo-GHMMs and phylo-SCFGs.) The phylo-HMM idea was first introduced for genome analysis by Churchill and Felsenstein, and further developed e.g. for RNA structure prediction by Knudsen and Hein. The use of phylo-grammars by Siepel, Pedersen, Bejerano, Haussler et al for gene prediction, evolutionary analysis of rate variation, and other forms of genome annotation has gotten lots of attention in recent years.

Much of the appeal of phylo-grammars is the straight transfer of intuition and expertise from the areas of HMMs and SCFGs. However, the well-known EM algorithms used to train these models (Baum-Welch, Inside-Outside) are a little less straightforward to apply to phylo- grammars. In contrast to (say) Baum-Welch, the phylo-EM algorithm is pretty hairy and not something you'd really want to implement twice.

In the past 4 years we have taken the theory of phylo-EM algorithms from a theoretical treatment (Holmes & Rubin, JMB, 2002) up to a full- blown open-source implementation of a general phylo-grammar prototyping, training and annotation engine (XRATE). Grammars can be specified using a Scheme-like format, "trained" on alignments using phylo-EM, and then used to annotate alignments. The phylo-EM code in our open-source C++ library can also be linked to by external applications (e.g. Jakob Pedersen's EVOFOLD program, which has been used to investigated recently-evolving human ncRNAs). Several developers have contributed full-time to the process, and there is considerable stability, including a battery of automated tests.

XRATE is an easy-to-use Unix app that brings the unrestricted power of phylo-grammars in reach of a first-year grad student or smart undergrad. In a historical aside, XRATE has its roots in a grammar compiler, TELEGRAPH, that was itself based on Ewan Birney's DYNAMITE (and is related to Guy Slater's EXONERATE). TELEGRAPH was presented at BOSC 2000. At BOSC 2006, I'll show how far XRATE has come by giving a tour of its rate-measurement and annotation abilities, accompanied by visualizations of the interesting variety of patterns (covariation, neighbor-dependence, conservation, lineage-specific acceleration, selection...) that can be observed in the mutation rates of genomic features.

Twease

Matt Wood, Ph. D. Cornell University.

Friday August 4th, 11:30am

With 16 million abstracts available in Medline, most searches match more documents than is possible to read. We asked if sentence level searches would be more effective in retrieving articles of interest than the whole abstract method currently used to support most biomedical searches. We present and evaluate a web-based tool to search Medline at the sentence level (available from http://www.twease.org/). The tool indexes each sentence of Medline individually and provides features that help correct for the lack of context introduced when searching sentences separately from the rest of the sentences in an abstract. We evaluated Twease with queries assembled from an independently obtained protein-protein interaction dataset (2,789 distinct interactions), as a measure of performance when retrieving abstracts with conjunctive queries of biomedical interest. Experimental results indicate that, on average, the first 25 Twease hits contain as many relevant abstracts as the first 100 PubMed hits. The first 25 Twease hits are also twice more likely to contain a relevant hit than the first 100 PubMed hits. These results indicate that sentence level searches, as implemented in Twease, are a competitive strategy when searching the biomedical literature for articles about multiple concepts (e.g., protein-protein interactions, or disease gene/protein relationships). Because a Twease index can be created directly from a text collection and does not require custom semantic resources, the approach implemented in Twease can be used to index and search any text collection of abstracts or full text articles. Twease implements highly scalable algorithms and approaches that will be discussed during the presentation and is released under the GNU General Public License. The distribution can be downloaded from: http://icb.med.cornell.edu/crt/twease/index.xml.

ZMap

Roy Storey, Sanger Institute, Hinxton, Cambridge

Friday August 4th, 1:30pm

We present a software package, ZMap, that is a visualisation tool for genomic features. The software is a multi-threaded application written in C, utilising the gnome toolkit (GTK2) and draws features on the foocanvas. ZMap accepts input from multiple sources in multiple formats across multiple genomes and is written in way that the addition of further formats is made as trivial as possible. Currently the list of formats includes GFF v2 and DAS1 which may reside in any one of; a file, an acedb instance, a http server. Multiple genomes and their associated features can be displayed in a single view as aligned blocks providing support for comparative annotation.

A wide complement of browser features are implemented in the ZMap GUI. Users may zoom to any resolution in the range from individual bases to a whole chromosome. The display may be split in both vertical and horizontal axis multiple times in a way akin to emacs. The contents of the display are completely copied in order that, as for emacs, the two views are independently scrollable. In this way it is possible to utilise screen space to view both ends of features at a much higher resolution than would otherwise be possible. Users are able to perform operations on the full sequence context, such as reverse complement, print, export and search.

Internally ZMap has a client/server architecture, where the GUI control thread acts as the client making requests to each server that communicates with a unique source. Each server runs within its own thread enabling the graphical thread to remain responsive. Various servers may add to the display at any point during the lifetime of the application. The features are merged with the current context allowing features from multiple sources to be viewed along side each other. As threads are separated from the control interface a user can kill the request in the event of an unresponsive or slow server.

ZMap does not natively include any utility for editing the features which it displays. It does however provide a powerful external interface with which to modify the features which are displayed on the canvas. ZMap utilises the X11 atom mechanism for interprocess communication, via which it is possible to communicate in xml. A perl module X11:XRemote is provided to facilitate ease of integration with perl annotation tools. Using this interface Sanger's production annotation tool otterlace is used to annotate sequence present in the Otter database which in turn updates to the Vega website.

Software License ----------------

The GNU General Public License (GPL) Version 2

CaIntegrator

Subha Madhavan, Ph.D. , National Cancer Institute

Saturday August 5th, 10:00am

Progress in finding better therapies for cancer treatment is hampered by the lack of opportunity to integrate biomedical data from disparate sources to enable translation of therapies from bench to bedside and back. Hence, a critical factor in the advancement of biomedical research and delivery is the ease with which data from clinical trials can be integrated, redistributed and analyzed both within and across functional domains. Novel biomedical informatics infrastructure and tools are essential for developing individualized patient treatment regimens based on the specific genomic signatures in each patientsed gene expression, Copy number Abnormality (CNA), SNPs, clinical trials data etc.) in a cohesive fashion.

Following are some of the high-level features of the caIntegrator framework:

- N-Tiered Architecture: The caIntegrator framework is implemented using Java 2 Enterprise Edition, a Data Warehouse, and various other open source technologies. It is designed to conform to an n- tiered architecture that includes several layers: A web-based graphical user interface, a business object layer, server components that process the queries and result sets, a data access layer and a data warehouse.

- A Common set of interfaces (APIs) and a domain object model: The domain object model called CGOM (Clinical Genomics Object Model) provides standardized programmatic access to the integrated biomedical data collected in the caIntegrator data system. The model represents data from clinical trials, microarray-based gene expression, SNP genotyping and copy number experiments, and Immunohistochemistry-based protein assays. Clinical domain objects in CGOM allow access to Clinical trial protocol, treatment arms, patient information, sample histology, clinical observations and assessments. Genomic domain objects allow access to biospecimen information, raw experimental data, in-silico transformation and analyses performed on the raw experimental datasets and biomarker findings. The application's user interface communicates with its caIntegrator-based middle-tier services via domain as well as business objects.

- A real-time analytical service that provides on-the-fly computational analysis capability for caIntegrator applications and currently supports class comparison analysis, principal component analysis and clustering analysis. It is designed to easily incorporate other types of analysis in the future, and scale to provide performance.

- The caIntegrator data system consists of a star schema database, which contains the clinical, and annotation data as dimensions, pre-calculated gene expression copy number data as facts. For performance reasons, normalized gene expression data used by the real time analysis module is stored as R-binary files.

- A plotting interface to allow visualization of genomic data (copy number scatter plots and ideogram) via the webGenome application (http://webgenome.nci.nih.gov).

The overall goal of the caIntegrator project is to provide a caBIG-compatible (https://cabig.nci.nih.gov/guidelines_documentation) framework with the infrastructural components needed to develop enterprise level translational applications. One such reference implementation at NCICB is REMBRANDT (Repository of Molecular BRAin Neoplasia DaTa) - http://rembrandt.nci.nih.gov. REMBRANDT is a powerful and intuitive informatics system designed to integrate genetic and clinical information from brain tumor clinical trials for improved research, disease diagnosis, and treatment. It provides researchers with the ability to perform ad hoc querying and reporting across multiple data domains, such as Gene Expression, Chromosomal aberrations and Clinical data. Scientists are able to answer basic questions related to a patient or patient population and view the integrated data sets in a variety of contexts. Tools that link data to other annotations such as cellular pathways, gene ontology terms and genomic information are embedded within this system. Two other cancer study applications that are being developed using the caIntegrator framework are:

- I-SPY Breast cancer trial: The primary object of the I SPY Trial is to identify surrogate markers of response to neoadjuvant chemotherapy that are predictive of pathologic remissions and survival in Stage III breast cancer. Goal is to identify Molecular markers and/or MRI results that predict 3-year disease-free survival in these patients.

- CGEMS: Cancer Genetic Markers of Susceptibility (CGEMS) is an NCI project to identify genetic alterations that make people susceptible to prostate and breast cancers. Goal of CGEMS is to analyze the entire genome for most common type of SNPs that are associated with each of these diseases.

caIntegrator knowledge framework offers a paradigm for rapid sharing of information and accelerates the process of analyzing results from various biomedical studies with the ultimate goal to rapidly change routine patient care.

Resources:

CaIntegrator website: http://caintegrator.nci.nih.gov REMBRANDT (caIntegrator reference implementation): http://rembrandt.nci.nih.gov REMBRANDT application URL: http://rembrandt-db.nci.nih.gov CaIntegrator open source license: http://ncicb.nci.nih.gov/download/caintegratorlicenseagreement.jsp

SHOGUN