- Open Access
Phenotypic and genotypic data integration and exploration through a web-service architecture
© Nuzzo et al; licensee BioMed Central Ltd. 2009
- Published: 15 October 2009
Linking genotypic and phenotypic information is one of the greatest challenges of current genetics research. The definition of an Information Technology infrastructure to support this kind of studies, and in particular studies aimed at the analysis of complex traits, which require the definition of multifaceted phenotypes and the integration genotypic information to discover the most prevalent diseases, is a paradigmatic goal of Biomedical Informatics. This paper describes the use of Information Technology methods and tools to develop a system for the management, inspection and integration of phenotypic and genotypic data.
We present the design and architecture of the Phenotype Miner, a software system able to flexibly manage phenotypic information, and its extended functionalities to retrieve genotype information from external repositories and to relate it to phenotypic data. For this purpose we developed a module to allow customized data upload by the user and a SOAP-based communications layer to retrieve data from existing biomedical knowledge management tools. In this paper we also demonstrate the system functionality by an example application of the system in which we analyze two related genomic datasets.
In this paper we show how a comprehensive, integrated and automated workbench for genotype and phenotype integration can facilitate and improve the hypothesis generation process underlying modern genetic studies.
- Data Warehouse
- Phenotype Definition
- Star Schema
- Genomic Knowledge
- Soap Request
One of the most challenging goals of current biomedical research is to link the genotypic and phenotypic information generated by high-throughput experimental technologies . The shift from hypothesis based to hypothesis-free research that has been made possible by these technological advances opens unprecedented new opportunities for studying biological systems on a large scale, at a low cost, and with a holistic perspective that promises to expand our understanding of biological processes and of their connections with clinically relevant outcomes. The price to pay for this paradigmatic shift is that researchers will increasingly need to handle very large volumes of heterogeneous data, both generated by their own experiments and retrieved from publicly available repositories of genomic knowledge. Integration, exploration, manipulation and interpretation of such data therefore need to become as automated as possible. The "traditional" data inspection and analysis methods are quickly becoming inadequate in a scenario in which an investigator can sample hundreds of thousands of variables in parallel: not only ad-hoc analysis methods need to be developed in order to address problems related with the statistical significance of the analysis results, but all phases of the scientific discovery process (hypothesis generation and testing, background knowledge gathering, experiment design, interpretation of results, generation of new knowledge) will have to be adapted to this new reality. In an era in which an entire new genome can be sequenced and annotated in a matter of days, it will become essential to be able to automatically link new observations and findings to pre-existing knowledge. Finally, new data storage and retrieval systems will need to be adopted in order to handle the unprecedented volumes of data and information being generated in an efficient and productive way. These tools, however, should be easily accessible to the broad research community, facilitating the discovery process by providing high usability and effective automation. Information Technology will therefore play an increasingly crucial role in modern biomedical and translational research , by developing the methods and tools that will allow researchers to bridge the gap between biomedical research and clinical applications. The ability to effectively address the challenges outlined above will have a direct, dramatic impact on the speed, accuracy and effectiveness of the scientific progress in all areas of the life sciences.
We therefore propose the application of data warehouse concepts to facilitate the investigation of biomedical data by researchers lacking technical expertise and database skills . Our system, called Phenotype Miner, provides a simple and effective way to organize, represent and navigate phenotypic data along multiple dimensions, and to select subsets of subjects based on one or more phenotypes of interest. Our current aim is to turn the system into a general tool for hypothesis generation, experiment design, automated annotation and biomedical data management, by relating phenotypic data to genomic knowledge. In this paper we describe the overall architecture of the system, which exploits a distributed architecture based on Web Services to integrate the Phenotype miner with two additional modules that support automated hypothesis generation process as an integral part of modern translational research. The first one is the Data Uploader, a tool to parse user-provided data sets of phenotypic data and to store them in the Phenotype Miner's database. The Data Uploader modular architecture makes it possible to easily customize both the parsing method (to support different data formats and representations) and the way in which data are stored in the database (to adapt them to the most appropriate structure for the specific analysis being performed). The second module allows the system to access external resources providing background genomic information and knowledge through a standard Web Services protocol. This tool can be used, for example, to automatically link the phenotypes under study with available genotype data on the basis of pre-existing knowledge about the relationship between the phenotypes and genomic markers.
In the following we present a detailed description of the system's architecture and usage, we illustrate the most important technological and methodological solutions we adopted for its implementation, and we present an application example. Using two sample datasets, the first one containing a dozen heterogeneous clinical measurements on about one hundred individuals and the second one containing their genotypes obtained with a SNP microarray, we demonstrate the system's usability and we highlight the ways in which it can be used to answer complex questions about genotype-phenotype correlations.
The Phenotype Miner, the core module of the system, was developed as an application of data warehousing to the domain of genetic studies. These studies rely on the integration and manipulation of large amounts of heterogeneous data, including genotypic, phenotypic and genealogical information, and therefore pose significant challenges related with structuring the data and querying them effectively. In particular, we showed how the definition and interpretation of phenotypic data can be improved through a multidimensional analysis approach .
Data Warehouse techniques: the OLAP engine for data exploration
While a normalized structure (i.e, one based on the Entity-Relation model) may be preferred for a correct management of the database in terms of data integrity and reliability, the adoption of query-oriented models that reflect the logical structure of the data elements greatly facilitates the task of filtering the data on the basis of the desired combination of phenotypes and patient features. Thus, we used a logical database design technique to support end-user queries in a data warehouse called "dimensional modelling", and we adopted a data structure called "star-join schema", that has become a standard for data warehouse applications. Unlike the Entity-Relation model, a dimensional model is very asymmetric. There is one large dominant table in the center of the schema, called fact table. It is the only table in the schema which is connected to the other tables with multiple joins. Such other tables, called dimensional tables, only require a single join to be referenced by the fact table .
Typically a clinical database can be modeled by a star schema in which each record in the fact table represents a combination of a clinical measure and its values on a specific date for a specific patient. Therefore, the dimensions are individuals, measurement time and measurement values: all of them can be further specified using a snowflake model, that is, a model in which a given dimension has relationships to other attributes of the same dimension (used to re-normalize complex dimensions in order to eliminate redundancy) .
Multidimensional analysis is implemented by software tools called OLAP (Online Analytical Processing) engines. Unlike Online Transaction Processing (OLTP), where typical operations read and modify individual and small numbers of records, OLAP engines deal with large quantities of data in real time, and operations are generally read-only. The term "online" implies that even though huge quantities of data are involved – typically many millions of records, occupying several gigabytes – the system must respond to queries fast enough to allow an interactive exploration of the data.
Moreover, formalization of the phenotype definition is needed to implement automated query generation. The definition and use of a formal phenotype definition allowed us to implement an automatic query generation tool, suitable for users who may not possess the necessary technical skills in query languages and database manipulation.
The Data Uploader
Phenotypic datasets are, in general, highly complex and deeply structured: the definition of the value types and ranges of the individual data elements, of the relationships that exist among them, and of the role they play is a function of the study being performed, and the structure of the database that will hold them should reflect the specific purpose for which they are going to be used. Moreover, phenotypic data comes in a very wide variety of formats, encodings and representations. It is therefore impossible to develop a single, universal method to import phenotypic data into a general-purpose tool such as the Phenotype Miner. Instead, we developed a modular component, the Data Uploader, which acts as the interface between the user-provided data source and the database. As described below, the Data Uploader can easily be reconfigured to adapt to changes in the database structure or to accept different input data formats. In this way, the Phenotype Miner can be used as an off-the shelf data mining tool, which is independent of the original data source used.
The SOAP interface
The final component of the system is a module to communicate with external resources through a suitable Web Services interface. According to the W3C, a Web Service is defined as "a software system designed to support interoperable machine to machine interaction over a network" . Web Services are Web-based communication protocols that allow a program to access one or more services available on a remote system over the network. The protocols specify how to define, locate, implement, and invoke services. In the case of SOAP , the most widely used protocol in the Web Services context, messages are encoded as XML documents and are transmitted between client and server using the HTTP/HTTPS protocols.
We developed a SOAP client interface by which the Phenotype Miner can access external repositories of genomic knowledge that support the Web Services protocols. In particular, we implemented methods to interact with Genephony, an online tool for genomic dataset annotation . In the current prototype, we exploit the functionalities provided by Genephony to automatically find SNPs related to a Mendelian phenotype by searching the OMIM database .
Thanks to the addition of the two above-described modules, the Phenotype Miner becomes a complete, general and flexible tool to integrate phenotypic and genomic information. In the most common usage scenario, the user imports patient phenotypic data into the database using a suitable Data Uploader and uses the Phenotype Miner components (Phenotype Editor, OLAP Engine, PedLauncher) to perform the desired selection and filtering operations on the dataset. The user may then use the available external knowledge bases to integrate the phenotypic data with genomic information in order to help in their interpretation or correlate them with available data at the genomic level. Finally, the resulting dataset can be browsed or exported for subsequent analysis.
The phenotype editor and the OLAP engine
Final users can create phenotype definitions using a graphical wizard developed in JAVA programming language. The tool runs as a Java Web Start application in order to exploit both the development facilities provided by stand-alone GUI APIs and to keep it available as a Web tool (instead of downloading and running it as a separate application). It interacts with two sections of the non-normalized database, i.e. the star schema of the clinical data and the phenotypes definition tables. The conditions (attributes and their values) may be defined by combo boxes, which provide lists of attributes according to the measures table of the star schema, suggesting the admissible ranges of values for each attribute. Once the rules are defined, the corresponding SQL string is created by merging conditions by AND operators, in order to: i) store the rules in the phenotype section tables, and ii) select the subgroup of individuals satisfying the conditions and storing the individuals-phenotype relationship. In the same way, it is possible to use another graphical panel to select previously defined phenotypes to be merged together by OR operators to create a new phenotype, which is then saved in the phenotypes section of the database.
When the OLAP engine starts, it reads an XML file containing the data definition. The first page of the system shows a set of check boxes containing the fields of the underlying tables, so that the user may choose the variable to be investigated (the phenotypes are among them). Once the features have been chosen, the engine loads information related to the individuals having that phenotype. Then a visual inspection of the measurement values can be performed expanding or collapsing cells of the resulting table, so that the analysis can be executed at different levels of detail.
The Data Uploader
The Data Uploader module allows the user to upload study data from text files directly into the system's database. It is composed of different modules (called "datasheet parsers" in Figure 3), each of which is designed to parse a different type of file. The Uploader engine parses the supplied file using the appropriate datasheet parser, and inserts the resulting data into the database according to the star-schema structure. User intervention is only needed to describe the logical connections between the fields in the source files and the data elements represented in the database. We assume that the data have already been pre-processed to ensure their quality, as data cleaning and filtering are beyond the scope of this paper.
The final result is that the data are automatically stored in the data warehouse according to the star-join structure exploited by the Phenotype Miner.
It is important to note that data import is not necessarily limited to parsing local files. As the use of Web Services becomes more widespread, it is conceivable that a growing number of repositories of phenotypic data (e.g. clinical information systems) will make their data available to authorized software agents using standard interoperability protocols such as SOAP. In this case, it will be sufficient to implement a SOAP-based Data Uploader to give our system the ability to automatically acquire data from remote primary sources.
The SOAP interface
The SOAP Interface is a module that provides the Phenotype Miner with the ability to access genomic knowledge stored within a remote resource. While there are several different types of messaging patterns available in SOAP, the most common is the Remote Procedure Call (RPC) pattern, in which one network node (the client) sends a request message to another node (the server) invoking the execution of a specified function on a specified set of arguments. The server executes the function, and automatically sends a response message back to the client with the results of the computation.
The client requests the creation of a session, in which all subsequent processing will take place;
The client queries Genephony for all OMIM entries containing the term (or terms) identifying the phenotype of interest; Genephony stores the resulting set in the current session;
The client asks for all genomic regions referenced in the OMIM entries (ie, regions known to be associated with the phenotype), and then for all SNPs belonging to these regions;
Finally, the client retrieves the resulting list of SNP identifiers.
The actual processing takes place on the Genephony server, with several advantages: there is no need for a local annotation database or for methods to access multiple external resources, the communication between the client and the server can be optimized (since only the final result set is transmitted to the client, and not all the intermediate steps), and the computational load on the client application is minimized. It should be noted that although OMIM only provides information about Mendelian diseases, the process just described could be applied equally well to other sources of knowledge about genotype-phenotype correlations (eg GAD, the Genetic Association Database ).
As an example scenario of the system usage, here we consider a case-control genome-wide association study. Association studies aim to find statistically significant differences in the distribution of a set of markers between a group of individuals showing a trait of interest (the cases) and a group of unrelated individuals who do not exhibit the trait (the controls) [14, 15]. Among the several different kinds of association studies, genome-wide association studies (GWAS) rely on a set of genetic markers covering the whole genome . This strategy is motivated when there is little or no a priori information about the location of the genetic cause of the phenotype being studied. Although the power of a genome-wide association study is usually low, such studies are useful to pinpoint areas of the genome that may contain candidate genes, and to guide a subsequent, more targeted analysis step. As the costs of genotyping decreases and the number of known, well characterized genetic markers in the human genome increases, the genome-wide approach represents an increasingly cost-effective way of generating testable research hypotheses.
Genome-wide association studies typically rely on two kinds of datasets, one collecting clinical (i.e. phenotypic) measurements, and the other one storing the individuals' genotypic markers values. In our case we assume the genotype dataset to be the result of large-scale SNP genotyping, while the clinical dataset may include any kind of measurement or observation. Starting from these two sources of information, the user's objective is to identify a set of individuals sharing a phenotype of interest, to identify a set of SNPs known to be related to the phenotype, and finally to extract from the genotype database the allelic values of the SNPs for the previously identified individuals. These steps will be reflected in the following sequence of operations in the system:
upload the subjects' phenotype information (i.e. clinical measurements);
define the phenotypes of interest;
find individuals showing the defined phenotype;
find the set of genomic markers known to be related to the phenotype;
finally, retrieve and download genotypes for those individuals and markers only.
The next step is to retrieve a set of markers that are known to be related to the phenotypes. Through the automatic SOAP request to Genephony described above, the user can retrieve all the SNPs (if any) related to the OMIM entry associated with the phenotype under study. In this example, the dataset produced by Genephony in response to the query contains about 48,000 SNPs. Of course not all SNPs will be actually needed for the purposes of the study, since many SNPs will be redundant with each other because of their proximity, and, more importantly, not all of them will be represented on the genotyping microarray used in the analysis.
Therefore, a further processing step consists in filtering the resulting set of SNPs according to the specific genotyping platform used in the study, and to other desired properties (e.g., validated SNPs only). This step too can be performed through a SOAP request to a genomic annotation service such as Genephony.
Since the phenotype definition is different, the system will not only retrieve a different set of patients, but also a (possibly) different set of genetic markers. After performing association analysis in the two cases, it would therefore be possible to compare the resulting sets of associated markers directly, for example to determine if some of them are associated with both phenotypes, or to identify markers that are associated with only one of the two conditions.
Thanks to the ability to access external resources for genomic annotation, the system could then provide the user with detailed information about the genomic context of the markers thus identified (e.g. proximity to genes, biological pathways the affected genes belong to, etc). The system here described can therefore be used to generate hypothesis about candidate genetic factors potentially involved in disease development, through dynamic and powerful data exploration and annotation features.
In this paper we presented a web-based software system aimed at addressing the problem of managing and exploring heterogeneous phenotypic and genotypic information in an automated, integrated and easy to use application. After describing the main system, called Phenotype Miner, which allows a dynamical inspection of clinical data, we described in more detail a set of desirable system's functionalities that we developed in order to: i) allow the user to upload experimental data, and ii) retrieve genomic information from existing biological knowledge bases and integrate it with user-supplied data. The typical application scenario of our system is in the context of studies aimed at investigating genetic factors potentially underlying phenotypes of interest.
We integrated different methodologies to develop the various software components of our system. We used a logical formalization for phenotype definition, a powerful graphical tool to define phenotypes, a data warehouse approach for dynamic, multi-dimensional data investigation, and a "Web Services" architecture to access external sources of genomic data and knowledge. The modular nature of the system makes it suitable for the development of new applications that integrate clinical data with already available external services and information resources. Future developments will concern the inclusion of new sources of information, such as investigation results from the literature, and a post-processing phase to provide specific output data formats for genetic analysis software, as well as including some preliminary statistical analysis.
A prototype version of the system is freely available at the URL http://bioinfo.unipv.it/PhenotypeMiner. A pre-populated example database is provided for demonstration purposes.
This work was supported in part by FIRB project "ITALBIONET – Rete Italiana di Bioinformatica", funded by the Italian Ministry of University and Research, and by NIH grant R01 HL87681-01, "Genome-Wide Association Studies in Sickle Cell Anemia and in Centenarians."
This article has been published as part of BMC Bioinformatics Volume 10 Supplement 12, 2009: Bioinformatics Methods for Biomedical Complex System Applications. The full contents of the supplement are available online at http://www.biomedcentral.com/1471-2105/10?issue=S12.
- Botstein D, Risch N: Discovering genotypes underlying human phenotypes: past successes for mendelian disease, future approaches for complex disease. Nature Genetics 2003, 33: 228–237. 10.1038/ng1090View ArticlePubMedGoogle Scholar
- Martin-Sanchez F, Iakovidis I, Norager S, Maojo V, de Groen P, Lei J, Jones T, Abraham-Fuchs K, Apweiler R, et al.: Synergy between medical informatics and bioinformatics: facilitating genomic medicine for future health care. J Biomed Inform 2004, 37: 30–42. 10.1016/j.jbi.2003.09.003View ArticlePubMedGoogle Scholar
- Nuzzo A, Segagni D, Milani G, Rognoni C, Bellazzi R: A dynamic query system for supporting phenotype mining in genetic studies. Stud Health Technol Inform 2007, 129(Pt 2):1275–1279.PubMedGoogle Scholar
- Kinball R, Ross M: The data warehouse toolkit. second edition. Wiley and Sons; 2002.Google Scholar
- Wyderka KC: Data warehouse technique for outcomes management. Health Management Technology 1999, 20: 16–17.PubMedGoogle Scholar
- Pentaho Analysis Service:The Mondrian OLAP Project. [http://mondrian.pentaho.org]
- Agarwala R, Biesecker LG, Hopkins KA, Francomano CA, Schaffer AA: Software for constructing and verifying pedigrees within large genealogies and an application to the old order amish of lancaster county. Genome Res 1998, 8: 211–221.View ArticlePubMedGoogle Scholar
- Sala C, Bione S, Crocco L, Gatti M, Poggiali E, Bellazzi R, Buetti I, Rognoni C, Camaschella C, Toniolo D: The val borbera project: epidemiological and genealogical analysis of an isolated population in northern Italy. Proceedings of the European Society of Human Genetics Conference 2006.Google Scholar
- W3C Working Group[http://www.w3.org/TR/soap/]
- W3C Working Group Note 11 February 2004. WebServices Glossary [http://www.w3.org/TR/ws-gloss/]
- Nuzzo A, Riva A: A knowledge management tool for translational research. Proceedings of the 1st AMIA Summit on Translational Bioinformatics 2008, 171.Google Scholar
- Online Mendelian Inheritance in Man, OMIM (TM): McKusick-Nathans Institute of Genetic Medicine, Johns Hopkins University (Baltimore, MD) and National Center for Biotechnology Information, National Library of Medicine (Bethesda, MD).[http://www.ncbi.nlm.nih.gov/omim/]
- Becker KG, Barnes KC, Bright TJ, Wang SA: The genetic association database. Nat Genet 2004, 36(5):431–432. 10.1038/ng0504-431View ArticlePubMedGoogle Scholar
- Balding DJ: A tutorial on statistical methods for population association studies. Nat Rev Genet 2006, 7(10):781–791. 10.1038/nrg1916View ArticlePubMedGoogle Scholar
- Cordell JC, Clayton DG: Genetic association studies. Lancet 2005, 366(9491):1121–1131. 10.1016/S0140-6736(05)67424-7View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.