Analysing the scalability of multiobjective evolutionary algorithms when solving the motif discovery problem

Abstract

In this paper we analyse the scalability of seven multiobjective evolutionary algorithms when they solve large instances of a known biological problem, the motif discovery problem (MDP). The selected algorithms are a population-based and a trajectory-based algorithms (DEPT and MO-VNS, respectively), three swarm intelligence algorithms (MOABC, MO-FA, and MO-GSA), a genetic algorithm (NSGA-II), and SPEA2. The MDP is one of the most important sequence analysis problems related to discover common patterns, motifs, in DNA sequences. A motif is a nucleic acid sequence pattern that has some biological significance as being DNA binding sites for a regulatory protein, i.e., a transcription factor (TF). A biologically relevant motif must have a certain length, be found in many sequences, and present a high similarity among the subsequences which compose it. These three goals are in conflict with each other, therefore a multiobjective approach is a good way of facing the MDP. In addition, in recent years, scientists are decoding genomes of many organisms, increasing the computational workload of the algorithms. Therefore, we need algorithms that are able to deal with these new large DNA instances. The obtained experimental results suggest that MOABC and MO-FA are the algorithms with the best scalability behaviours.