Early modeling of preterm birth risk using a biocollection of pregnancy samples created at different gestational stages according to the "longitudinal biobanking" strategy

Aim : to develop an algorithm for early modeling of high risk of preterm birth (PB) using a biocollection of pregnancy samples created according to the "longitudinal biobanking" strategy, based on clinical, proteomic, and transcriptomic data. Materials and Methods . A prospective cohort study was conducted. Serum and plasma samples were retrieved dynamically from a biocollection during the first and second trimesters of gestation from 18 pregnant women with PB high risk developed cervical insufficiency (CI), and from 18 pregnant women with PB low risk not complicated by СI at any stage of gestation. Clinical and anamnestic predictors were analyzed, and serum relaxin and elastin levels, as well as marker microRNAs (hsa-miR-432-5p, hsa-miR-134-5p, hsa-miR-431-5p, hsa-miR-122-5, and hsa-miR-34a-5p) were quantitated. Results . Based on the most significant clinical and anamnestic predictors, a proteomic predictor (serum relaxin level), and transcriptomic predictors (marker microRNAs expression) using a support vector machine (SVM), a predictive model was developed, which high efficiency was confirmed by ROC curves (AUC = 0.83 during training and 0.7 in cross-validation). This allows the developed classifier to be used as a tool for assessing PB risk. Conclusion . Biocollections of pregnant women's samples, formed according to the "longitudinal biobanking" strategy, can serve as an effective platform for developing technologies for early predicted PB risk of obstetric complications.