Li Rujun, Wang Haotian, Yu Qiunan, Cai Jing, Jiang Liangzhen, Luo Ximei, Zou Quan, Lv Zhibin
College of Biomedical Engineering, Sichuan University, Chengdu 610041, China.
College of Food and Biological Engineering, Chengdu University, Chengdu 610106, China.
Foods. 2025 Jun 6;14(12):2014. doi: 10.3390/foods14122014.
Antioxidant peptides (AOPs) have the natural properties of food preservatives; they are capable of improving the oxidation stability of food while also providing additional benefits such as disease prevention. Traditional experimental methods for identifying antioxidant peptides are time consuming and costly, so effective machine learning models are increasingly being valued by researchers. In this study, we integrated amino acid composition, transformation, and distribution (CTD) and block substitution matrix 62 (BLOSUM62) to develop an SVM-based AOP prediction model called AOPxSVM. This strategy significantly improves the prediction accuracy of the model by comparing 15 feature combinations and feature selection strategies, with their effectiveness being visually verified using UMAP. AOPxSVM achieves high accuracy values of 0.9092 and 0.9330, as well as Matthew's correlation coefficients (MCCs) of 0.8253 and 0.8670, on two independent test sets, both surpassing the state-of-the-art methods based on the same test sets, thus demonstrating AOPs' excellent identification capability. We believe that AOPxSVM can serve as a powerful tool for identifying AOPs.
抗氧化肽（AOPs）具有食品防腐剂的天然特性；它们能够提高食品的氧化稳定性，同时还能提供诸如疾病预防等额外益处。传统的抗氧化肽鉴定实验方法既耗时又昂贵，因此有效的机器学习模型越来越受到研究人员的重视。在本研究中，我们整合了氨基酸组成、转化和分布（CTD）以及块替换矩阵62（BLOSUM62），以开发一种基于支持向量机的AOP预测模型，称为AOPxSVM。通过比较15种特征组合和特征选择策略，该策略显著提高了模型的预测准确率，其有效性通过UMAP进行了直观验证。AOPxSVM在两个独立测试集上分别实现了0.9092和0.9330的高精度值，以及0.8253和0.8670的马修斯相关系数（MCCs），均超过了基于相同测试集的现有最佳方法，从而证明了AOPs出色的识别能力。我们相信AOPxSVM可以作为识别AOPs的有力工具。
