Integrating information retrieval, summarization, natural language processing, and user interfacing for phenotype-genotype association.

摘要

Due to the large number of articles published in the biomedical domain, researchers often spend a lot of time finding information relevant to their research. For this dissertation, we have developed and integrated three applications that extract relevant information from text -- BioRover, FigureSearch and SimGenes. BioRover takes a gene or disease name as input and generated a list of related disease or genes. Related genes and diseases are extracted by identifying gene and disease named entities from over 600 million sentences. A dictionary-based disease name tagger was developed to identify disease entities and gene names were identified using a machine-learning based tagger. BioRover also allows users to filter results based on sentence modalities negation and speculation, and genetic and epigenetic factors such as mutation, methylation, and phosphorylation. Negation and speculation are identified using a machine learning-based tagger. FigureSearch is a search engine for published biomedical figures. FigureSearch indexes over 16 million figures from over 5 million full-text articles. For each figure, a four-sentence summary is generated by extracting sentences from the full-text of the article. The summary contains one sentence classified as Introduction, one classified as Methods, one classified as Results and one classified as Discussion. A machine learning-based classifier was used to classify sentences. SimGenes identifies semantically similar genes for a reference gene.;The Gene Ontology annotations are used to calculate semantic similarity between genes. A web application was developed to allow users to use these systems. We believe that these applications will help access relevant information quickly and using SimGenes, researchers can generate new hypothesis. All three systems are available online as free web-applications -- BioRover -- http://biorover.askhermes.org/, FigureSearch -- http://figuresearch.askhermes.org and SimGenes -- http://simgenes.askhermes.org.