Decoding preeclampsia: A fusion of multi-view machine learning and multi-omics to identify putative inflammation-related mechanisms

Preeclampsia (PE) is a leading cause of maternal and fetal morbidity and mortality worldwide, with placental inflammation recognized as a central pathogenic feature, yet the upstream triggers and inflammatory mechanisms remain incompletely understood. Here, we combined placental single-cell transcriptomics with gut metagenomic and metabolomic profiling to characterize inflammatory signatures in PE. Stratified analyses across clinical subgroups-defined by fetal number, onset timing, and fetal sex-revealed that placental single-cell transcriptomics coupled with multi-view machine learning consistently prioritized bacteria-associated inflammatory features across all subgroups. Superimposed on this shared foundation, we identified subgroup-specific trajectories: twin PE exhibited IL-1-dominant inflammation with compensatory antioxidant metabolic shifts, while singleton PE showed IFN-II-associated immune activation. Early-onset PE displayed sexual dimorphism-male fetuses featured bacterial defense pathways, lipid metabolic programs, and trophoblast-confined glycolysis, while female fetuses exhibited angiogenesis, chemotaxis, nitric oxide signaling pathways, and glycolytic reprogramming in immune cells, whereas late-onset PE exhibited comparatively attenuated inflammatory activity. Gut metagenomic profiling revealed enrichment of lipopolysaccharide (LPS)-producing taxa and depletion of beneficial commensals in PE, accompanied by metabolomic alterations that aligned with inflammatory pathways also highlighted in placental analyses. Collectively, these findings reveal a conserved bacteria-associated inflammatory program in PE that is modulated by clinical context and linked to gut microbial dysbiosis.