Related Experiment Video
Updated: Sep 13, 2025
11:02
Identification and Quantification of Deranged Metabolites in Critically Ill Patients Using NMR-Based Metabolomics
Published on: November 29, 2024
649
Comparative analysis of machine learning techniques in metabolomic-based preterm birth prediction.
Ying-Chieh Han1, Jane Shearer1,2,3, Chunlong Mu2,3
1Department of Biomedical Engineering, Faculty of Engineering, University of Calgary, 2500 University Dr. NW, Calgary, AB T2N 1N4, Canada.
Computational and Structural Biotechnology Journal
|August 1, 2025
Summary
Machine learning models can predict preterm birth using serum metabolomics. XGBoost with resampling showed the highest accuracy, identifying specific metabolites and metabolic pathways linked to preterm delivery.
Area of Science:
- Life Science Research
- Computational Biology
- Biochemistry
Background:
- Machine learning (ML) is increasingly used in life sciences.
- This study evaluated ML models for predicting preterm birth.
- Serum metabolomics data from the third trimester was utilized.
Purpose of the Study:
- To assess the efficacy of various ML models in predicting preterm birth.
- To identify key metabolites and metabolic pathways associated with preterm delivery.
- To optimize predictive accuracy in small-scale clinical datasets.
Main Methods:
- Utilized serum samples from 48 preterm and 102 term delivery mothers.
- Applied four ML algorithms: PLS-DA, logistic regression, ANN, and XGBoost.
- Evaluated models using confusion matrices, AUROC, and feature importance (SHAP).
Main Results:
- Non-linear models (ANN, XGBoost) showed marginal improvement over linear models (PLS-DA, logistic regression).
- XGBoost with bootstrap resampling achieved the highest performance (AUROC = 0.85).
- Acylcarnitines, amino acid derivatives, and disrupted tyrosine/phenylalanine/tryptophan metabolism were key discriminators.
Conclusions:
- Metabolomics-based preterm birth prediction is complex.
- An iterative, model-driven approach is recommended for small datasets.
- ML models, particularly XGBoost, show promise for preterm birth prediction.
