AUTOMATED EXTRACTION OF METADATA FROM EXPERIMENT DESIGNS

DESCRIPTION (provided by applicant): The querying of biological databases by biomedical researchers is hampered by the absence of metadata, which describe the "why" and "how" of the data generated by an experiment. Consequently, it is difficult for researchers to query for experiments stored in these databases, decreasing the effective value of this hard won knowledge. We propose to alleviate this deficiency by developing an innovative metadata extraction software tool, ShowMe, to automatically infer the goal of an experiment from a graphical description of the experiment design. Using ShowMe, a researcher will draw the design using a graphical template and a set of icons. These icons are either pre-defined or user-defined, and associated to ShowMe's ontological classes. Once a diagram has been generated, ShowMe will deconstruct it and infer the experiment's goal by analyzing the ontological attributes of the diagram. With the researcher's review and approval, ShowMe will then submit these metadata to a database repository, to be associated with the data produced by that experiment. In this way, ShowMe will enable metadata-driven database queries by alleviating the obstacle of requiring researchers to explicitly state experiment metadata. In this Phase I project, we will ascertain the feasibility of this approach using microarray experiment designs. Our specific aims are to (i) determine whether experiment diagrams created via ShowMe's Template accurately captures the design of microarray experiments; and (ii) evaluate whether ShowMe can correctly infer the goals of these experiments. In a future Phase II, we will further develop and commercialize ShowMe to support a broader array of experiment designs, such as those in high-throughput drug screening and in vivo pharmacology. Facilitating such metadata extraction will thus improve the queryability of heterogeneous data sources. The ultimate goal of the proposal is to facilitate the reuse of data from biomedical research's vast databases of experiment data by developing a tool for the automated capture of metadata.