Hierarchical Boosting for Gene Function Prediction

Functional classification of genes using diverse bio-molecular data obtained from high-throughput technologies is a fundamental problem in bioinformatics and functional genomics. Genes are organized and classified according to a hierarchical classification scheme and each gene will participate in multiple activities. Flat classifiers, that work on non-hierarchical classification problems independently, do not take into account the hierarchical structure of the functional class taxonomies. Therefore, they are not able to utilize the information inherent in the class hierarchy. Moreover, independent classifiers, where each classifier predicts the gene membership to a particular class, may lead to an inconsistent set of predictions for a hierarchically structured classification scheme. In this paper, we propose HMLBoosting algorithm for the problem of hierarchical multi-label classification in the context of gene function prediction. HML-Boosting exploits the hierarchical dependencies among the classes. Extensive experiments on four bio-molecular datasets using two approaches for class-membership inconsistency correction during the testing phase, the top-down approach and the bottom-up approach, show that HML-Boosting algorithm outperforms flat classifiers using different evaluation metrics. In addition, we carry out a detailed comparison of the two approaches for class-membership inconsistency correction during the testing phase.