Hierarchical multi-label classification for protein function prediction: A local approach based on neural networks

摘要: In Hierarchical Multi-Label Classification problems, each instance can be classified into two or more classes simultaneously, differently from conventional classification. Additionally, the are structured in a hierarchy, form of either tree directed acyclic graph. Hence, an assigned to paths hierarchical structure, resulting complex classification problem with possibly hundreds classes. Many methods have been proposed deal such some them employing single classifier all simultaneously (global methods), and others many classifiers decompose original set subproblems (local methods). this work, we propose novel local method named HMC-LMLP, which uses one Multi-Layer Perceptron per level. The predictions level used as inputs network responsible for next We make use distinct algorithms: Back-propagation Resilient Back-propagation. addition, error measure specially tailored multi-label problems training networks. Our is compared state-of-the-art algorithms, protein function prediction datasets. experimental results show that our approach presents competitive predictive accuracy, suggesting artificial neural networks constitute promising alternative biological data.