Identification and prediction of milk-derived bitter taste peptides based on peptidomics technology and machine learning method

• Published in: Food chemistry 2024-02, Vol.433, p.137288-137288, Article 137288
• Main Authors: Yu, Yang, Liu, Shengchi, Zhang, Xinchen, Yu, Wenhao, Pei, Xiaoyan, Liu, Li, Jin, Yan
• Format: Article
• Language: English
• Subjects: Bitter taste peptides; Calcium mobilization; CPM-BP; Human bitter taste receptor T2R4; Light Gradient Boosting Machine algorithm; Peptidomics
• ISSN: 0308-8146; 1873-7072
• DOI: 10.1016/j.foodchem.2023.137288

[Display omitted] •A LightGBM model was built to predict bitter peptides (BPs) with accuracy of 90.3 %.•An expanded dataset and a new group of BPs’ characteristic factors were proposed.•724 peptides were significantly different between fresh and spoiled UHT milk.•180 potential BPs in spoiled UHT milk were predicted, in which three were verified.•Combining peptidomics with machine learning provides effective ways to find BPs. Bitter taste peptides (BPs) are vital for drug and nutrition research, but large-scale screening of them is still time-consuming and costly. This study developed a complete workflow for screening BPs based on peptidomics technology and machine learning method. Using an expanded dataset and a new combination of BPs’ characteristic factors, a novel classification prediction model (CPM-BP) based on the Light Gradient Boosting Machine algorithm was constructed with an accuracy of 90.3 % for predicting BPs. Among 724 significantly different peptides between spoiled and fresh UHT milk, 180 potential BPs were predicted using CPM-BP and eleven of them were previously reported. One known BP (FALPQYLK) and three predicted potential BPs (FALPQYL, FFVAPFPEVFGKE, EMPFPKYP) were verified by determination of calcium mobilization of HEK293T cells expressing human bitter taste receptor T2R4 (hT2R4). Three potential BPs could activate the hT2R4 and are demonstrated to be BPs, which proved the effectiveness of CPM-BP.