Umami-MRNN: Deep learning-based prediction of umami peptide using RNN and MLP

• Published in: Food chemistry 2023-03, Vol.405, p.134935, Article 134935
• Main Authors: Qi, Lulu, Du, Jialuo, Sun, Yue, Xiong, Yongzhao, Zhao, Xinyao, Pan, Daodong, Zhi, Yueru, Dang, Yali, Gao, Xinchang
• Format: Article
• Language: English
• Subjects: Deep learning; Multi-layer perceptron; Predictor; Recurrent neural network; Umami peptides
• ISSN: 0308-8146; 1873-7072
• DOI: 10.1016/j.foodchem.2022.134935

•This article is the first to use neural network to predict the taste of peptides.•Compared with other common machine learning methods, Umami-MRNN is more suitable for dealing with Classification of umami peptides.•The prediction results of Umami-MRNN through 10-fold cross-validation tests and independent tests demonstrate the model’s robust and excelling performance. Umami components are an important part of food condiments, and the use of umami peptides in the condiment industry has received great attention. However, traditional methods for umami peptide identification are time-consuming, labor-intensive, and difficult to achieve high throughput. Therefore, it is essential to develop an effective algorithm to identify potential umami peptides. In this study, we proposed a prediction method for umami peptides called Umami-MRNN. We constructed a merged model for the Multi-layer Perceptron and Recurrent Neural Network. We then developed predictors with six feature vectors as the input. We trained the neural networks using the training dataset and selected hyperparameters of machine learning models via a 10-fold cross-validation. The independent tests showed that Umami-MRNN achieved an accuracy of 90.5% and a Matthews correlation coefficient value of 0.811. To assist the scientific community, we also developed a publicly accessible web server at https://umami-mrnn.herokuapp.com/.