Automated Functional Annotation of Maize Genes Using Artificial Neural Networks for Literature Analysis
Dr. Jon Beck, Dr. Diane Janick-Buckner, and Dr. Brent Buckner, Faculty Mentors
Microarray hybridization experiments have produced large amounts of raw data for biological analysis. In many cases, though, there is too much data for humans to meaningfully use. Biologists at Truman State University have spent three years manually functionally annotating 6,000 of 38,000 Zea mays genes which are spotted onto a series of microarray chips. The remaining 32,000 genes on these chips are still unannotated. We have developed an artificial neural network (ANN) system to annotate these genes. The training and testing sets were composed of 6,000 manual annotations. A BLAST search of an unannotated sequence results in a series of hits and associated scientific publications. Inputs for the ANNs represent the presence or absence of specific keywords in the abstracts of these publications. A keyword vocabulary was iteratively developed to analyze the publication abstracts. To test prediction accuracy, double-blind tests and statistical analyses were performed.
Keywords: Zea mays, artificial neural network, literature, annotation, gene function, BLAST
Presentation Type: Oral Paper
Location: VH 1428