Figure 1.

An ensemble framework based on the SVM that integrates diverse datasets in the context of GO hierarchy. After pre-processing the data, we developed an approach that consists of an ensemble of three different classifiers: 1, a single SVM classifier for each GO term was trained on combined data; 2, single SVM classifiers were combined through Bayesian networks to correct their predictions based on the hierarchical relationship between GO terms in the GO directed acyclic graph; and 3, a naïve Bayes classifier was built for each GO term to directly integrate the results of single-dataset SVM classifiers. The bootstrap held-out values on the training set were used to characterize each classifier's performance, and the ensemble prediction was formed by selecting the best performing classifier on each GO term. GO, Gene Ontology; SVM, support vector machine.