Early detection of gestational diabetes and preeclampsia through human maternal serum spectral profiling: a machine learning combined ATR-FTIR approach

Gestational diabetes mellitus (GDM) and preeclampsia (PE) are concerning pregnancy-related disorders since they are among the leading causes of maternal and fetal morbidity and mortality worldwide. Unfortunately, despite their clinical relevance, they are usually diagnosed only in the second or even third trimester of pregnancy, when pathological processes are already established, and preventive interventions are limited. Therefore, the need for new tools able to improve early diagnosis appears fundamental. In this study, a multidisciplinary approach was applied on serum samples collected from pregnant women with diagnosis of GDM and PE, with the aim to identify reliable and non-invasive biomarkers for improving the early diagnosis of the two pathologies. Data collected by Attenuated Total Reflectance - Fourier Transform InfraRed (ATR-FTIR) were processed using multivariate and machine learning approaches, including Partial Least Squares–Discriminant Analysis (PLS-DA), K-Nearest Neighbors (KNN), and Random Forest (RF). This spectral-statistical analysis revealed, in both pathologies, spectral profiles distinct from those of healthy samples, with alterations mainly in protein structure, lipid metabolism, and carbohydrate content. Overall, these results enabled the identification of specific spectral markers for GDM and PE and demonstrated the potential translational value of this approach. Overall, this proof-of-concept study demonstrates that ATR-FTIR spectroscopy combined with machine learning represents a promising, rapid, and cost-effective strategy for early maternal serum profiling, supporting timely risk stratification in pregnancy care.