Deep neural network (DNN) is a useful machine learning approach, although its applicability to metabolomics studies has rarely been explored. Here we describe the development of an ensemble DNN (EDNN) algorithm and its applicability to metabolomics studies. As a model case, the developed EDNN approach was applied to metabolomics data of various fish species collected from Japan coastal and estuarine environments for evaluation of a regression performance compared with conventional DNN, random forest, and support vector machine algorithms. This study also revealed that the metabolic profiles of fish muscles were correlated with fish size (growth) in a species-dependent manner. The performance of EDNN regression for fish size based on metabolic profiles was superior to that of DNN, random forest, and support vector machine algorithms. The EDNN approach, therefore, should be helpful for analyses of regression and concerns pertaining to classification in metabolomics studies.

中文翻译：

---

集成深度神经网络在代谢组学研究中的应用

深度神经网络 (DNN) 是一种有用的机器学习方法，尽管很少探索其在代谢组学研究中的适用性。在这里，我们描述了集成 DNN (EDNN) 算法的开发及其在代谢组学研究中的适用性。作为模型案例，开发的 EDNN 方法应用于从日本沿海和河口环境收集的各种鱼类的代谢组学数据，以评估与传统 DNN、随机森林和支持向量机算法相比的回归性能。这项研究还表明，鱼类肌肉的代谢特征以物种依赖性方式与鱼类大小（生长）相关。基于代谢特征的鱼大小的 EDNN 回归性能优于 DNN、随机森林和支持向量机算法。