A Machine Learning Model Based on First-Trimester Lipidomic Signatures for Predicting Metabolic Pregnancy...

Area of Science:

• Biochemistry
• Genomics and Proteomics
• Computational Biology

Background:

• Gestational diabetes mellitus (GDM) and macrosomia significantly impact maternal and neonatal health.
• Early molecular dysregulations precede clinical manifestation of these pregnancy complications.
• Predictive biomarkers are needed for timely intervention and personalized pregnancy management.

Purpose of the Study:

• To develop early predictive models for GDM and macrosomia using first-trimester serum lipidomic signatures.
• To identify novel lipid biomarkers associated with GDM and macrosomia.
• To evaluate the efficacy of machine learning models in predicting these pregnancy complications.

Main Methods:

• A case-control study involving 119 women during first-trimester screening.
• Serum lipidomic profiling using shotgun mass spectrometry.
• Machine learning models (Random Forest, XGBoost) with 10-fold cross-validation after Shapley value-based feature selection.

Main Results:

• Identified potential GDM biomarkers: elevated triacylglycerol (TG) 55:7, decreased 13-Docosenamide, plasmenyl-phosphatidylcholine (PC P)-36:2, and phosphatidylcholine (PC) 42:7.
• Identified potential macrosomia biomarkers: phosphatidylglycerol (PG) (i-, a- 29:0), 4-Hydroxybutyric acid, and Pantothenol.
• GDM prediction model achieved 87% sensitivity and 89% specificity; macrosomia model achieved 87% sensitivity and 93% specificity.

Conclusions:

• First-trimester serum lipidomics combined with clinical data and machine learning can accurately predict GDM and macrosomia risk.
• This integrated approach shows promise for a clinical tool for early intervention and personalized pregnancy care.
• Early identification of high-risk pregnancies facilitates timely management to improve maternal and neonatal outcomes.