Application of Molecular Transformer approach for predicting the potential reactions to generate advanced glycation end products in infant formula

• Published in: Food chemistry Vol. 407; p. 135143
• Main Authors: Yang, Huihui, Bai, Xiaosen, Feng, Baolong, Wang, Qinghua, Meng, Li, Wang, Fengzhong, Wang, Yutang
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01.05.2023
• Subjects: Advanced glycation end products reaction; Food, Processed; Glycation End Products, Advanced - chemistry; Humans; Infant; Infant formula; Infant Formula - chemistry; Machine learning; Prospective Studies; Transformer; Database; Transformer; Advanced glycation end products reaction; Infant formula; Machine learning
• ISSN: 0308-8146; 1873-7072
• DOI: 10.1016/j.foodchem.2022.135143

•AGEs chemical reaction molecular database was established.•Transformer models were established for different datasets.•Prediction of AGEs in infant formula with the established optimal model. Advanced glycation end products (AGEs) are associated with the occurrence of human chronic diseases, and exist commonly in thermally processed foods, such as infant formula. Existing research mainly focuses on the discrete simulation system, which is time-consuming and challenging, but accumulates of a large amount of valuable data. This study aimed to propose a specific Molecular Transformer-based model trained on the data curated from literature to predict the chemical reaction of AGEs, and apply it to infant formula to observe which new reactions could generate AGEs. The model achieved top-3 accuracy of 76.0% on the total dataset. Based on the model prediction results, five reactions were selected for experimental verification, and four of them were consistent with the model prediction results. This prospective study might potentially revolutionize the discovery of AGEs reactions and provide theoretical guidelines for designing a safer infant formula.