Machine learning approach identifies meconium metabolites as potential biomarkers of neonatal hyperbilirubinemia

Shujuan Zeng; Zhangxing Wang; Peng Zhang; Zhaoqing Yin; Xunbin Huang; Xisheng Tang; Lindong Shi; Kaiping Guo; Ting Liu; Mingbang Wang; Huixian Qiu

doi:10.1016/j.csbj.2022.03.039

Machine learning approach identifies meconium metabolites as potential biomarkers of neonatal hyperbilirubinemia

Comput Struct Biotechnol J. 2022 Apr 2:20:1778-1784. doi: 10.1016/j.csbj.2022.03.039. eCollection 2022.

Authors

Shujuan Zeng¹, Zhangxing Wang², Peng Zhang², Zhaoqing Yin³, Xunbin Huang¹, Xisheng Tang⁴, Lindong Shi¹, Kaiping Guo¹, Ting Liu¹, Mingbang Wang^{5

6}, Huixian Qiu¹

Affiliations

¹ Division of Neonatology, Longgang District Central Hospital of Shenzhen, Guangdong 518116, China.
² Division of Neonatology, Shenzhen Longhua People's Hospital, Guangdong 518109, China.
³ Division of Neonatology, The People's Hospital of Dehong Autonomous Prefecture, Mangshi, Yunnan 678400, China.
⁴ Oncology Department, Longgang District Central Hospital of Shenzhen, Shenzhen 518116, China.
⁵ Shanghai Key Laboratory of Birth Defects, Division of Neonatology, Children's Hospital of Fudan University, National Center for Children's Health, Shanghai 201102, China.
⁶ Microbiome Therapy Center, South China Hospital, Health Science Center, Shenzhen University, Shenzhen 518116, China.

Abstract

Background: The gut microbiota plays an important role in the early stages of human life. Our previous study showed that the abundance of intestinal flora involved in galactose metabolism was altered and correlated with increased serum bilirubin levels in children with jaundice. We conducted the present study to systematically evaluate alterations in the meconium metabolome of neonates with jaundice and search for metabolic markers associated with neonatal jaundice.

Methods: We included 68 neonates with neonatal hyperbilirubinemia, also known as neonatal jaundice (NJ) and 68 matched healthy controls (HC), collected meconium samples from them at birth, and performed metabolomic analysis via liquid chromatography-mass spectrometry.

Results: Gut metabolites enabled clearly distinguishing the neonatal jaundice (NJ) and healthy control (HC) groups. We also identified the compositions of the gut metabolites that differed significantly between the NJ and HC groups; these differentially significant metabolites were enriched in aminyl tRNA biosynthesis; pantothenic acid and coenzyme biosynthesis; and the valine, leucine and isoleucine biosynthesis pathways. Gut branched-chain amino acid (BCAA) levels were positively correlated with serum bilirubin levels, and the area under the receiver operating characteristic curve of the random forest classifier model based on BCAAs, proline, methionine, phenylalanine and total bilirubin reached 96.9%, showing good potential for diagnostic applications. Machine learning-based causal inference analysis revealed the causal effect of BCAAs on serum total bilirubin and NJ.

Conclusions: Altered gut metabolites in neonates with jaundice showed that increased BCAAs and total serum bilirubin were positively correlated. BCAAs proline, methionine, phenylalanine are potential biomarkers of NJ.

Keywords: AUROC, the area under the ROC; BCAA, branched-chain amino acid; Gut microbiota; HC, healthy controls; KEGG, Kyoto Encyclopedia of Genes and Genomes; LC-MS, liquid chromatography-mass spectrometry; MSUD, maple syrup urine disease; Machine learning; NJ, neonatal jaundice; OPLS-DA, orthogonal partial least squares-discriminant analysis; PCA, the principal component analysis; PLS, partial least-squares regression; ROC, receiver operating characteristic; branched-chain amino acid; causal inference; metabolome; neonatal hyperbilirubinemia.