Phylogenetic inference allows building a hypothesis about the evolutionary relationships between a group of species, which is usually represented as a tree. The phylogenetic inference problem can be seen as an optimization problem, searching for the most qualified tree among all the possible topologies according to a selected criterion. These criteria can be based on different principles. Due to the combinatorial number of possible topologies, diverse heuristics and meta-heuristics have been proposed to find approximated solutions according to one criterion. However, these methods may result in several phylogeny trees which could be in conflict with one another. In order to deal with this problem, models based on multiobjective optimization with different configurations have been used. In this paper, we propose an ad-hoc multiobjective memetic algorithm (MO-MA) to infer phylogeny using two objectives: 1) maximum parsimony and 2) likelihood. Several population operators and local search strategies are proposed and evaluated in order to measure their contribution to the algorithm. Additionally, we perform a comparison among different configurations and tree rearrangement strategies. The results show that the proposed MO-MA is able to identify a Pareto set of solutions that include new trees which were nondominated by solutions from the current state of the art single-objective optimization tools. Furthermore, the MO-MA improves the results presented in the literature for multiobjective approaches in all of the studied data sets. These results make our proposal a good alternative for phylogenetic inference.

中文翻译：

---

基于NSGA-II方案的系统发育树推理模因算法

系统发育推断允许建立关于一组物种之间进化关系的假设，这通常表示为一棵树。系统发育推理问题可以看作是一个优化问题，根据选定的标准在所有可能的拓扑中搜索最合格的树。这些标准可以基于不同的原则。由于可能拓扑的组合数量，已经提出了多种启发式和元启发式以根据一种标准找到近似解。然而，这些方法可能会导致几种可能相互冲突的系统发育树。为了解决这个问题，已经使用了基于具有不同配置的多目标优化的模型。在本文中，我们提出了一种临时多目标模因算法 (MO-MA) 来使用两个目标来推断系统发育：1) 最大简约性和 2) 似然性。提出并评估了几种种群算子和局部搜索策略，以衡量它们对算法的贡献。此外，我们对不同的配置和树重排策略进行了比较。结果表明，所提出的 MO-MA 能够识别帕累托解决方案集，其中包括不受当前最先进单目标优化工具解决方案支配的新树。此外，MO-MA 改进了所有研究数据集中的多目标方法文献中提出的结果。这些结果使我们的提议成为系统发育推断的一个很好的选择。