Tracing Endometriosis: Coupling deeply phenotyped, single-cell based Endometrial Differences and AI for disease pathology and prediction

Abstract

AbstractEndometriosis, affecting 1 in 9 women, presents treatment and diagnostic challenges. To address these issues, we generated the biggest single-cell atlas of endometrial tissue to date, comprising 466,371 cells from 35 endometriosis and 25 non-endometriosis patients without exogenous hormonal treatment. Detailed analysis reveals significant gene expression changes and altered receptor-ligand interactions present in the endometrium of endometriosis patients, including increased inflammation, adhesion, proliferation, cell survival, and angiogenesis in various cell types. These alterations may enhance endometriosis lesion formation and offer novel therapeutic targets. Using ScaiVision, we developed neural network models predicting endometriosis of varying disease severity (median AUC = 0.83), including an 11-gene signature-based model (median AUC = 0.83) for hypothesis-generation without external validation. In conclusion, our findings illuminate numerous pathway and ligand-receptor changes in the endometrium of endometriosis patients, offering insights into pathophysiology, targets for novel treatments, and diagnostic models for enhanced outcomes in endometriosis management.