From gene expression to pregnancy prediction: towards precision ART through systems biology and Bayesian modeling

Purpose: This study investigates molecular differences between receptive and non-receptive endometrium by analyzing RNA cargo of uterine fluid-derived extracellular vesicles (UF-EVs). A total of 82 women who underwent blastocyst transfer with confirmed euploid embryos via Pre-implantation Genetic Testing (PGT-A) were included and followed to assess clinical pregnancy outcomes. The aim is to identify the transcriptomic profile of the endometrium during the window of implantation (WOI) by comparing gene expression between patients who achieved pregnancy and those who did not, and to use the WOI gene expression profile to predict pregnancy outcomes through a Bayesian logistic model. Methods: UF samples were collected on day 7 after the urinary LH surge (LH+7) in the cycle preceding the scheduled blastocyst transfer. In the subsequent natural cycle (LH+7), an ultrasound-guided transfer of the euploid blastocyst was performed, and clinical outcomes were recorded. A Differential Gene Expression (DGE) analysis (pregnant vs. not pregnant) was conducted, followed by a gene co-expression network analysis (WGCNA), which identified modules of co-expressed genes. These gene modules were integrated with clinical data to develop a Bayesian logistic regression model. Results: UF-EVs collected from 82 women were analyzed. Among the patients, 37 achieved pregnancy, while 45 did not. Differential gene expression analysis identified 966 significant genes (nominal p < 0.05), with a predominant upregulation in the pregnant group. Applying SEQC thresholds (nominal p < 0.01, |log₂FC| > 1) refined the list to 262 differentially expressed genes, the majority of which were overexpressed in patients who conceived. Four genes,RPL10P9, LINC00621, MTND6P4, and LINC00205, remained significant after multiple testing correction. Gene Set Enrichment Analysis (GSEA) revealed enriched biological processes such as adaptive immune response, lymphocyte-mediated immunity, and ion homeostasis. Weighted Gene Co-expression Network Analysis (WGCNA) grouped differentially expressed genes into four modules, with the grey and brown modules showing the strongest positive correlation with pregnancy. Functional enrichment of these modules highlighted immune regulation, transcriptional activity, and extracellular matrix remodeling. A Bayesian logistic regression model integrating gene expression modules with clinical variables accurately predicted pregnancy outcomes with 82.9% accuracy, an F1-score of 0.80, and an AUC of 0.88. Positive associations were observed for the grey module and co-expression patterns in the brown and turquoise modules, while increased EVs size and a history of miscarriages were negatively associated with pregnancy. Conclusions: This study demonstrates that UF-EV transcriptomic profiling can non-invasively capture endometrial receptivity signatures during the WOI. The integration of molecular and clinical data enables accurate pregnancy prediction, offering a promising approach to personalize embryo transfer timing and improve outcomes in ART.