Trie-based Apriori Motif Discovery Approach

Abstract

One of the hardest and long-standing problems in Bioinformatics is the problem of motif discovery in biological sequences. It is the problem of finding recurring patterns in these sequences. Apriori is a well-known data mining algorithm. It is used to mine frequent patterns in large datasets. In this paper, we would like to apply Apriori to the common motifs discovery problem. We propose three modifications so that we can adapt the classic Apriori to our problem. First, the Trie data structure is used to store all biological sequences under examination. Second, both of the frequent pattern extraction and the candidate generation steps are done using the same data structure, the Trie . The Trie allows to simultaneously search all possible starting points in the sequence for any occurrence of the given pattern. Third, instead of using only the support as a measure to assess frequent patterns, a new measure, the normalized information content (normIC), is proposed which is able to distinguish motifs in real promoter sequences. Preliminary experiments are conducted on Tompa's benchmark to investigate the performance of our proposed algorithm, the Trie-based Apriori Motif Discovery (TrieAMD). Results show that our algorithm outperforms all of the tested tools on real datasets for average sensitivity.