Wang Peiyao, Chen Peichun, Yang Xinjie, Cen Ziyan, Zhang Yu, He Qimin, Wu Benqing, Huang Xinwen
Department of Genetics and Metabolism, Children's Hospital of Zhejiang University School of Medicine, National Clinical Research Center for Child Health, Hangzhou, China.
Department of Preventive Health Care, Shenzhen Guangming Maternity and Child Care Hospital, Shenzhen, China.
Clin Transl Med. 2025 Sep;15(9):e70467. doi: 10.1002/ctm2.70467.
This study aimed to characterise urinary organic acid profiles in Neonatal Intrahepatic Cholestasis caused by Citrin Deficiency (NICCD) and develop a diagnosis model to distinguish NICCD patients from those in the non-specific metabolic abnormalities group (NAG), both of which exhibit elevated urinary 4-hydroxyphenyllactic acid (4-HPLA) and 4-hydroxyphenylpyruvic acid (4-HPPA), potentially leading to misdiagnosis.
A retrospective study was conducted from February 2021 to February 2025, enrolling 105 NICCD patients, 144 healthy controls (HC), and 298 individuals from NAG. Urine organic acids were measured using gas chromatography-mass spectrometry. Data from NICCD and NAG collected before October 2024 were used for model training and internal testing, with later data serving as an external validation. A three-step feature selection strategy identified biomarkers. Five machine learning (ML) methods were used to construct the model. Performance was compared using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1 score, etc. RESULTS: Compared to HC, NICCD patients exhibited 39 differential metabolites, enriched in tyrosine, aspartate, pyruvate, lipoic acid, and TCA cycle pathways. 4-HPLA, 4-HPPA, galactitol, 4-hydroxyphenylacetic acid, pyruvic acid, quinolinic acid, homovanillic acid, 4-hydroxybenzoic acid, and malic acid showed high diagnostic performance (AUC > .8). Nine robust markers were identified between NICCD and NAG. The random forest model demonstrated superior classification performance, with high AUC, accuracy, F1 score, and low Brier score. An online calculator was developed for clinical use.
Our findings highlight NICCD metabolic enrichment in energy and amino acid pathways and present an interpretable ML model for distinguishing NICCD from those of NAG.
本研究旨在描述因瓜氨酸缺乏引起的新生儿肝内胆汁淤积症（NICCD）患者的尿有机酸谱，并建立一个诊断模型，以区分NICCD患者与非特异性代谢异常组（NAG）患者，这两组患者均表现出尿中4-羟基苯乳酸（4-HPLA）和4-羟基苯丙酮酸（4-HPPA）升高，可能导致误诊。
2021年2月至2025年2月进行了一项回顾性研究，纳入105例NICCD患者、144例健康对照（HC）和298例NAG个体。采用气相色谱-质谱法测定尿有机酸。2024年10月之前收集的NICCD和NAG数据用于模型训练和内部测试，后期数据用作外部验证。采用三步特征选择策略确定生物标志物。使用五种机器学习（ML）方法构建模型。使用受试者操作特征曲线下面积（AUC）、准确性、敏感性、特异性、F1分数等比较模型性能。结果：与HC相比，NICCD患者表现出39种差异代谢物，富集于酪氨酸、天冬氨酸、丙酮酸、硫辛酸和三羧酸循环途径。4-HPLA、4-HPPA、半乳糖醇、4-羟基苯乙酸、丙酮酸、喹啉酸、高香草酸、4-羟基苯甲酸和苹果酸显示出较高的诊断性能（AUC>.8）。在NICCD和NAG之间确定了9个稳健的标志物。随机森林模型表现出卓越的分类性能，具有高AUC、准确性、F1分数和低布里尔分数。开发了一个在线计算器供临床使用。
我们的研究结果突出了NICCD在能量和氨基酸途径中的代谢富集，并提出了一个可解释的ML模型，用于区分NICCD与NAG。
