INTRODUCTION With chronic kidney disease (CKD), kidney becomes damaged overtime and fails to clean blood. Around 15% of US adults have CKD and nine in ten adults with CKD do not know they have it. OBJECTIVE Early prediction and accurate monitoring of CKD may improve care and decrease the frequent progression to end-stage renal disease. There is an urgent demand to discover specific biomarkers that allow for monitoring of early-stage CKD, and response to treatment. METHOD To discover such biomarkers, shotgun high throughput was applied to the detection of serum metabolites biomarker discovery for early stages of CKD from 703 participants. Ultra performance liquid chromatography coupled with high-definition mass spectrometry (UPLC-HDMS)-based metabolomics was used for the determination of 703 fasting serum samples from five stages of CKD patients and age-matched healthy controls. RESULTS AND CONCLUSION We discovered a set of metabolite biomarkers using a series of classic and neural network based machine learning techniques. This set of metabolites can separate early CKD stage patents from normal subjects with high accuracy. Our study illustrates the power of machine learning methods in metabolite biomarker study.

中文翻译：

---

机器学习蒸馏代谢物生物标志物用于早期肾损伤。

简介对于慢性肾脏病（CKD），肾脏会随着时间的流逝而受损，无法清洁血液。在美国约有15％的成年人患有CKD，十分之九的CKD成年人不知道自己患有该病。目的CKD的早期预测和准确监测可以改善护理水平，并减少至终末期肾脏疾病的频繁进展。迫切需要发现特定的生物标记物，以监测早期CKD以及对治疗的反应。方法为发现此类生物标志物，将gun弹枪高通量用于检测703名参与者的CKD早期阶段的血清代谢物生物标志物发现。超高效液相色谱结合高分辨质谱（UPLC-HDMS）的代谢组学方法测定了来自五个阶段的CKD患者和年龄匹配的健康对照者的703个空腹血清样品。结果与结论我们使用一系列基于经典和神经网络的机器学习技术发现了一组代谢物生物标志物。这组代谢物可以将CKD早期阶段的专利与正常受试者高精度地分开。我们的研究说明了机器学习方法在代谢物生物标志物研究中的作用。这组代谢物可以将CKD早期阶段的专利与正常受试者高精度地分开。我们的研究说明了机器学习方法在代谢物生物标志物研究中的作用。这组代谢物可以将CKD早期阶段的专利与正常受试者高精度地分开。我们的研究说明了机器学习方法在代谢物生物标志物研究中的作用。