MMP3C v2: a network-based framework decoding metabolic plasticity in rheumatoid arthritis, enabling accurate...

Metabolic plasticity, the ability of cells to dynamically adapt their metabolic pathways in response to changing environments, is a hallmark of rheumatoid arthritis (RA) pathogenesis and plays a critical role in immune dysfunction. However, scalable methods to quantify inter-pathway crosstalk in RA remain lacking. To address this gap, we present MMP3C v2, an updated network-based framework that integrates gene expression with protein-protein interaction network topology to compute directed pairwise metabolic plasticity (PMP) scores. We applied MMP3C v2 to ~3400 bulk transcriptomes (RA, osteoarthritis, systemic lupus erythematosus, and healthy controls) and ~228 000 single-cell transcriptomics from blood and synovium to profile RA-associated PMP alterations and develop diagnostic classifiers. We found that a single PMP-derived signature demonstrated strong predictive capability for diagnosis. Then, we developed a feature selection pipeline and combined it with 110 machine learning model combinations, by which we established the optimal ensemble classifier (stepwise forward selection + ridge regression), achieving robust and generalized performance (mean area under the curve (AUC) = 0.935; mean F1 score = 0.915) across 12 independent validation cohorts, outperforming seven previously published models. Single-cell analysis revealed cell-type-specific PMP remodeling: a Warburg-like shift in synovial macrophages (↑glycolysis, ↑pentose phosphate pathway, ↓oxidative phosphorylation). Cell-cell communication analysis highlighted FN1-centered signaling linked to glucose metabolic remodeling in myofibroblasts. Collectively, MMP3C v2 establishes metabolic pathway crosstalk as a core diagnostic feature of RA, enabling interpretable and cross-platform diagnostic modeling and the identification of cell-type-specific PMP patterns. The open-source R package mmp3c supports reproducible analysis and broad application.