Learning pathogenic proteins across fractured and heterogeneous data

In the following work, we test a generalized approach to integrating, transforming and learning data from disparate data sources for the classification of bacterial proteins involved in pathogenesis. We rely on the implicit inter-linkages between biological databases to draw relevant records, and leverage statistical learning methods to infer classification based on abundant, albeit noisy, data. Results suggest that types of public biological information have varying degrees of effectiveness in predictive data mining.