Metaheurísticas, optimización multiobjetivo y paralelismo para descubrir motifs en secuencias de ADN

Abstract

The resolution of complex problems by using evolutionary algorithms is one of the most researched issues in Computer Science. The main goal of this thesis is directly related with the development of new algorithms that can solve this kind of problems with the least possible computational time, improving the results achieved by the existing methods. To this end, we combine three important concepts: metaheuristics, multiobjective optimization, and parallelism. For doing this, we first look for a significant optimization problem that had not been solved in an efficient way and we find the Motif Discovery Problem (MDP). MDP aims to discover over-represented short patterns (motifs) in a set of DNA sequences that may have some biological significance. To address it, we defined a multiobjective formulation adjusted to the real-world biological requirements, we implemented a total of ten algorithms of different nature (population, trajectory, collective intelligence...), analyzing aspects such as the ability to scale and converge. Finally, we designed parallel techniques, by using parallel and distributed programming environments as OpenMP and MPI, which try to combine the properties of several metaheuristics in a single application. The obtained results are discussed in detail through numerous statistical tests, and the achieved predictions are compared with those discovered by a total of thirteen well-known biological tools. The drawn conclusions demonstrate that using multiobjective optimization in metaheuristic techniques favors the discovery of quality solutions, and that parallelism is useful for combining the properties of different evolutionary algorithms.