A screening test proposal for congenital defects based on maternal serum metabolomics profile

Area of Science:

• Biochemistry
• Genetics
• Machine Learning

Background:

• Noninvasive prenatal testing historically identifies <50% of chromosomal anomalies.
• Ultrasound diagnoses congenital defects later in pregnancy.
• Metabolomic analysis offers a novel, noninvasive early screening approach for fetal malformations.

Purpose of the Study:

• To evaluate the diagnostic accuracy of an ensemble machine learning model using maternal serum metabolomic signatures.
• To detect fetal malformations, including chromosomal anomalies and structural defects.

Main Methods:

• A multicenter observational retrospective study with training and validation cohorts.
• Untargeted metabolomics analysis via gas chromatography-mass spectrometry.
• Ensemble machine learning model developed using weighted voting of 9 individual models.

Main Results:

• Significant metabolic differences identified between women with and without fetal malformations.
• The screening test achieved 99.4% accuracy in the validation cohort.
• High specificity (99.9%) and moderate sensitivity (78%) were observed.

Conclusions:

• Clinical validation of a metabolomics-based prenatal screening test for congenital defects.
• Further research is needed to identify specific malformation types.
• Larger study populations are required to confirm findings.