Abstract：
This study investigates the possibility of utilizing drip as a non-destructive method for assessing the freshness and spoilage of chicken meat. The quality parameters [pH, volatile base nitrogen (VBN), and total aerobic bacterial counts (TAB)] of chicken meat were evaluated over a 13-day storage period in vacuum packaging at 4 °C. Simultaneously, the metabolites in the chicken meat and its drip were measured by nuclear magnetic resonance. Correlation (Pearson\'s and Spearman\'s rank) and pathway analyses were conducted to select the metabolites for model training. Binary logistic regression (model 1 and model 2) and multiple linear regression models (model 3-1 and model 3-2) were trained using selected metabolites, and their performance was evaluated using receiver operating characteristic (ROC) curves. As a result, the chicken meat was spoiled after 7 days of storage, exceeding 20 mg/100 g VBN and 5.7 log CFU/g TAB. The correlation analysis identified one organic acid, eight free amino acids, and five nucleic acids as highly correlated with chicken meat and its drip during storage. Pathway analysis revealed tyrosine and purine metabolism as metabolic pathways highly correlated with spoilage. Based on these findings, specific metabolites were selected for model training: ATP, glutamine, hypoxanthine, IMP, tyrosine, and tyramine. To predict the freshness and spoilage of chicken meat, model 1, trained using tyramine, ATP, tyrosine, and IMP from chicken meat, achieved a 99.9 % accuracy and had an ROC value of 0.884 when validated using drip metabolites. This model 1 was improved by training with tyramine and IMP from both chicken meat and its drip (model 2), which increased the ROC value for drip metabolites from 0.884 to 0.997. Finally, selected two metabolites (tyramine and IMP) can predict TAB and VBN quantitatively through models 3-1 and 3-2, respectively. Therefore, the model developed using metabolic changes in drip demonstrated the capability to non-destructively predict the freshness and spoilage of chicken meat at 4 °C. To make generic predictions, it is necessary to expand the model\'s applicability to various conditions, such as different temperatures, and validate its performance across multiple chicken batches.
摘要：
这项研究调查了利用滴水作为评估鸡肉新鲜度和腐败的非破坏性方法的可能性。质量参数[pH，挥发性碱氮(VBN)，在4°C的真空包装中，在13天的储存期内评估鸡肉的总需氧细菌计数（TAB）]。同时,用核磁共振法测定鸡肉及其滴落物中的代谢产物。进行相关性(Pearson's和Spearman's秩)和途径分析以选择用于模型训练的代谢物。使用选定的代谢物对二元逻辑回归(模型1和模型2)和多元线性回归模型(模型3-1和模型3-2)进行训练,并使用受试者工作特性(ROC)曲线评估其性能。因此,鸡肉在储存7天后变质了，超过20mg/100gVBN和5.7logCFU/gTAB。相关性分析确定了一种有机酸，八种游离氨基酸,和五种核酸与鸡肉及其在储存过程中的滴落高度相关。途径分析显示酪氨酸和嘌呤代谢是与腐败高度相关的代谢途径。基于这些发现,选择特定的代谢物进行模型训练：ATP，谷氨酰胺,次黄嘌呤,IMP,酪氨酸,还有酪胺.为了预测鸡肉的新鲜度和腐败，模型1，使用酪胺训练，ATP,酪氨酸,和鸡肉的IMP，使用滴注代谢物验证时,准确率为99.9%,ROC值为0.884.该模型1通过用酪胺和来自鸡肉及其滴剂的IMP训练来改进(模型2)，将滴注代谢物的ROC值从0.884增加到0.997。最后,选择两种代谢物(酪胺和IMP)可以分别通过模型3-1和3-2定量预测TAB和VBN。因此,使用滴水中的代谢变化开发的模型证明了在4°C下非破坏性地预测鸡肉的新鲜度和腐败的能力。要进行通用预测，有必要将模型的适用性扩展到各种条件，例如不同的温度，并验证其在多个鸡肉批次中的性能。
