EndoTwin-W: glycodelin-A and CA-125 as non-invasive biomarkers of endometrial receptivity derived from a multiscale computational digital twin

Endometrial receptivity assessment currently requires invasive tissue biopsy, yet recent randomized trials have questioned the clinical utility of biopsy-based approaches. Here we present EndoTwin-W, a four-layer mechanistic computational model that simulates human endometrial remodeling from hormone inputs through receptor binding, pathway scoring, and continuous-time Markov chain cell-state transitions across 17 cell states. Transition rates were optimized against scRNA-seq and microarray data, then validated by 5-fold cross-validation on an independent bulk RNA-seq cohort (n=236 biopsies), achieving significant correlations for 16 of 17 cell states (mean Spearman r = 0.505) with benchmark dominance over three null models for 13 of 17 states. The model identifies glycodelin-A (PAEP) and CA-125 (MUC16) as mechanistically grounded candidate circulating biomarkers capturing two principal receptivity failure modes: inadequate decidualization and excessive inflammation. Hill-function prediction of serum glycodelin-A shows strong rank-order calibration (Spearman rho = 0.833, p = 0.010). Cross-condition held-out validation against 9 independent datasets (244 samples) achieves significant concordance in 5 of 9 datasets (median rho = 0.435). A cross-dataset receptivity index analysis across 18 GEO datasets (21 comparisons) demonstrates mean AUC = 0.599 with correct direction in 76% of analyses, including significant RNA-seq validation (AUC = 0.770, p = 0.003). The divergence between predicted and measured biomarker values defines a Progesterone Resistance Score quantifying decidualization deficit and inflammation burden. EndoTwin-W provides a mechanistic framework and candidate blood-based biomarkers for receptivity assessment; prospective paired serum-tissue validation is required before clinical use. Author SummaryAssessing whether the uterine lining is ready for embryo implantation usually requires an invasive biopsy that is costly and cannot be repeated every cycle. We built a computer model called EndoTwin-W that simulates how ovarian hormones reshape the endometrium through hormone receptors, intracellular signaling, and changing cell states across the menstrual cycle. When we tested the model against published gene-expression datasets from hundreds of patient samples, it matched known endometrial cell states in 16 of 17 categories. Our main finding is that two blood proteins, glycodelin-A and CA-125, may serve as non-invasive markers of receptivity. Glycodelin-A reflects decidualization; CA-125 reflects inflammation. When the models predictions disagree with measured blood levels, the mismatch defines a two-dimensional progesterone resistance score that may help explain why some patients do not respond to progesterone despite normal hormone levels. We provide an open research website (https://endotwin-w.com: mirror: https://endotwinw.com) for exploration, but prospective clinical studies are still needed before this approach could guide patient care.