Related Experiment Video
Updated: Mar 21, 2026
09:24
Functional Assessment of Intestinal Permeability and Neutrophil Transepithelial Migration in Mice using a Standardized Intestinal Loop Model
Published on: February 11, 2021
8.2K
A Biologically Informed Machine Learning Pipeline Uncovers Metabolic Features of Intestinal Barrier Dysfunction
Ke-Xin Liu1, Ze-Yuan Liang1, Tong Li1
1Pukou Hospital of Chinese Medicine Affiliated to China Pharmaceutical University, Department of Chinese Medicines Analysis, China Pharmaceutical University, Nanjing 210009, China.
Analytical Chemistry
|March 19, 2026
Summary
Machine learning identified key metabolic biomarkers for intestinal barrier dysfunction, reducing the need for large clinical studies. This approach aids in understanding complex diseases and evaluating therapeutic responses.
Area of Science:
- Biomedical Engineering
- Computational Biology
- Metabolomics
Background:
- Intestinal barrier dysfunction is implicated in various diseases but lacks clear diagnostic criteria.
- Biomarker discovery for this condition is challenging due to high costs and the need for large cohorts.
Purpose of the Study:
- To develop a machine learning (ML) pipeline integrating network biology and ensemble feature selection (EFS) for identifying metabolic features of intestinal barrier dysfunction.
- To establish a biologically interpretable and statistically rigorous framework for biomarker discovery in complex diseases.
Main Methods:
- Constructed a heterogeneous dataset from murine models and derived a continuous intestinal barrier index using a supervised regression task.
- Employed network biology-informed EFS to identify a stable subset of 10 core metabolites across multiple regression architectures.
- Validated prioritized metabolic features using independent clinical data and in vitro biotransformation assays.
Main Results:
- Identified a robust subset of 10 core metabolites consistently associated with intestinal barrier dysfunction (R²: 0.604-0.654).
- Confirmed the clinical applicability of key metabolites, including serotonin, N-acetylputrescine, d-phenylalanine, and hippuric acid.
- Demonstrated ML's utility in feature prioritization and hypothesis generation, reducing reliance on large-scale cohorts.
Conclusions:
- The developed ML pipeline effectively identifies metabolic features linked to intestinal barrier dysfunction, offering a cost-efficient approach to biomarker discovery.
- The findings provide novel insights into the metabolic underpinnings of intestinal barrier dysfunction and facilitate the evaluation of multicomponent therapeutic strategies.
- This integrated approach supports the management of complex multifactorial diseases by enabling efficient biomarker identification and validation.
