Data integration through heterogeneous ensembles for protein function prediction

Several data integration strategies have been developed to leverage the inherent diversity and complementarity of multi-modal datasets to build comprehensive predictive models about the biomedical processes and/or entities being studied. Here, we propose an effective approach, named Ensemble Integration (EI), to data integration and predictive modeling that assimilates local models inferred from these datasets into heterogenous ensembles. Through the sample challenging problem of protein function prediction, we demonstrate that EI performs better for this task than some other forms of data integration.