Proteomics data interoperation with applications to integrated data mining and enhanced information retrieval.

摘要

This thesis addresses the problem of data integration and interoperation of large-scale, widely distributed and independently maintained data, focusing on biological proteomics data which exemplifies the problem and has a practical need for better interoperation, and shows how such integrated data can be leveraged for important applications such as detailed cross-database queries in support of scientific exploratory data analysis and enhanced information retrieval. Semantic web RDF and RDF databases, which fit the problem well, are used to build two biological data integration systems called YeastHub and LinkHub. YeastHub is a lightweight semantic web data warehouse of joined RDF-formatted biological (yeast) data and allows RDF query access to it. LinkHub focuses on a high-level structuring principal or "scaffold" for biological data, storing biological identifiers (e.g. for proteins, genes, etc.) and the complex relationships among them as a large RDF directed labeled graph; LinkHub is used through web interactive and query interfaces and also complements YeastHub. Through several nontrivial RDF queries of the joined YeastHub and LinkHub data, we demonstrate that practical integrated biological data analysis can be achieved by basic, lightweight methods which don't attempt to solve the complete integration problem.; A key focus of the LinkHub system is support for enhanced information retrieval of web documents and articles from the biomedical scientific literature (PubMed). We attach documents to identifier nodes in the LinkHub RDF graph and provide for the flexible retrieval of the documents through queries of the RDF graph's relational structure. In addition, we use the LinkHub RDF relational data and attached documents as training sets to construct classifiers for document relevance ranking in support of enhanced automated information retrieval of web or biomedical scientific literature documents related to biological identifiers. The results of experiments done to empirically measure the performance of this enhanced automated information retrieval for proteomics (UniProt) identifier-related documents through the use of a manually curated bibliography of yeast protein-specific literature citations are presented.