Improvement of FP-Growth Algorithm for Mining Description-Oriented Rules

In the paper new modification of the rules induction method for description of gene groups using Gene Ontology based on FP-growth algorithm is proposed. The modification takes advantage of the hierarchical structure of GO graph, specific property of a single prefix-path FP tree and the fact that if we generate rules for description purposes we do not include into rule premise two GO terms that are in parent-children relation. The proposed algorithms was implemented and tested with two different expression datasets. Time performance of old and new approach is compared together with descriptions obtained with two methods. The results show that the new method allows generating rules faster, while the number of rules and coverage is similar in both approaches.