Machine learning based metabolomic and genetic profiles for predicting multiple brain phenotypes

Xueli Zhang, Yu Huang, Shunming Liu, Shuo Ma, Min Li, Zhuoting Zhu, Wei Wang, Xiayin Zhang, Jiahao Liu, Shulin Tang, Yijun Hu, Zongyuan Ge, Honghua Yu (Corresponding Author), Mingguang He, Xianwen Shang

Research output: Journal article publicationJournal articleAcademic researchpeer-review

Abstract

Background: It is unclear regarding the association between metabolomic state/genetic risk score(GRS) and brain volumes and how much of variance of brain volumes is attributable to metabolomic state or GRS. Methods: Our analysis included 8635 participants (52.5% females) aged 40–70 years at baseline from the UK Biobank. Metabolomic profiles were assessed using nuclear magnetic resonance at baseline (between 2006 and 2010). Brain volumes were measured using magnetic resonance imaging between 2014 and 2019. Machine learning was used to generate metabolomic state and GRS for each of 21 brain phenotypes. Results: Individuals in the top 20% of metabolomic state had 2.4–35.7% larger volumes of 21 individual brain phenotypes compared to those in the bottom 20% while the corresponding number for GRS ranged from 1.5 to 32.8%. The proportion of variance of brain volumes (R [2]) explained by the corresponding metabolomic state ranged from 2.2 to 19.4%, and the corresponding number for GRS ranged from 0.8 to 8.7%. Metabolomic state provided no or minimal additional prediction values of brain volumes to age and sex while GRS provided moderate additional prediction values (ranging from 0.8 to 8.8%). No significant interplay between metabolomic state and GRS was observed, but the association between metabolomic state and some regional brain volumes was stronger in men or younger individuals. Individual metabolomic profiles including lipids and fatty acids were strong predictors of brain volumes. Conclusions: In conclusion, metabolomic state is strongly associated with multiple brain volumes but provides minimal additional prediction value of brain volumes to age + sex. Although GRS is a weaker contributor to brain volumes than metabolomic state, it provides moderate additional prediction value of brain volumes to age + sex. Our findings suggest metabolomic state and GRS are important predictors for multiple brain phenotypes.
Original languageEnglish
Article number1098
Pages (from-to)1-13
Number of pages13
JournalJournal of Translational Medicine
Volume22
Issue number1
DOIs
Publication statusPublished - 3 Dec 2024

Keywords

  • Brain phenotype
  • Genetic risk score
  • Metabolomic profiles
  • Metabolomic state
  • Moderation analysis
  • Prediction value

ASJC Scopus subject areas

  • General Biochemistry,Genetics and Molecular Biology

Fingerprint

Dive into the research topics of 'Machine learning based metabolomic and genetic profiles for predicting multiple brain phenotypes'. Together they form a unique fingerprint.

Cite this