Integration of heterogeneous data sources for gene function prediction using decision templates and ensembles of learning machin