Identify potential circRNA-disease associations through a multi-objective evolutionary algorithm
Identify potential circRNA-disease associations through a multi-objective evolutionary algorithm
复制标题
DOI：
10.1016/j.ins.2023.119437
复制
发表时间：
2023-08
期刊：
影响因子：
--
通讯作者：
Yuchen Zhang;Xiu-juan Lei;Cai Dai;Yi Pan;Fangxiang Wu
中科院分区：
文献类型：
--
作者：
Yuchen Zhang;Xiu-juan Lei;Cai Dai;Yi Pan;Fangxiang Wu
越来越多的研究表明，CircRNAs由于其固有的稳定性，可以作为各种疾病的标志物。许多计算方法，特别是人工智能方法，已经被应用于预测CircRNA与疾病的关联。然而，这些预测方法的目标函数通常是单一的和标准的。单一的目标函数很难反映预测问题的特点，导致预测精度较低。对于CircRNA疾病预测问题，从来没有一种方法来设计一组目标函数，并使用智能优化算法来求解它。本文提出了一种通过多目标进化算法识别CircRNA疾病关联的ICDMOE方法。基于相似网络的矩阵分解和模块化，定义了解空间，设计了4个目标函数。采用一种改进的基于分解的多目标进化算法(MOEA/D)来求解这些目标函数。在MOEA/D中，提出了一种基于惩罚的自适应边界交叉策略，该策略既保证了收敛，又保证了Pareto前沿解的多样性。实验结果表明，ICDMOE比基于纯矩阵分解的方法、基于非自适应MOEA/D的方法以及其他预测方法具有更好的性能。此外，我们发现预测的疾病-CircRNAs可以被现有的研究、miRNA调控和表达谱等证实。这表明ICDMOE可以为生物医学实验提供良好的候选。
More and more studies have demonstrated that circRNAs can be used as markers of various diseases due to their inherent stability. Numerous computational methods, especially artificial intelligence approaches, have been applied to the prediction of circRNA-disease associations. However, the objective functions of these prediction methods are usually single and standard. The single objective function hardly reflects the characteristics of the prediction problem, and leads to low prediction accuracy. There has never been a method to design a set of objective functions for the circRNA-disease prediction problem and solve it by using intelligent optimization algorithms. In this paper, a method ICDMOE is proposed to identify circRNA-disease associations via a multi-objective evolutionary algorithm. A solution space is defined and four objective functions are designed based on matrix factorization and modularity of similarity networks. An improved decomposition-based multi-objective evolutionary algorithm (MOEA/D) is also employed to solve these objective functions. In the MOEA/D, an adaptive penalty-based boundary intersection strategy is proposed, which not only ensures convergence, but also guarantees the diversity of solutions in the Pareto front. Finally, the experimental results show that ICDMOE has better performance than pure matrix factorization-based methods, non-adaptive MOEA/D-based methods and other prediction methods. Furthermore, we find that the predicted disease-circRNAs can be confirmed by existing studies, miRNA regulations and expression profiles,etc. These indicate that ICDMOE can provide good candidates for biomedical experiments.