Longitudinal Transcriptomic Remodeling of Adipose Tissue After Bariatric Surgery Revealed by Differential Expression and Explainable Machine Learning

Abstract Bariatric surgery leads to major improvements in metabolic health, but the long‑term molecular changes in white adipose tissue (WAT) that support these benefits are not fully understood. We aimed to characterize how WAT gene expression remodels over time after Roux‑en‑Y gastric bypass (RYGB) and to identify stable markers of long‑term adaptation. We reanalyzed a public longitudinal transcriptomic dataset from women undergoing RYGB, with WAT biopsies collected before surgery and at 2 and 5 years after surgery. We combined traditional differential expression analysis with machine learning and SHAP‑based interpretation to identify biologically meaningful and reproducible gene signatures. Both approaches showed a clear shift in WAT gene‑expression profiles after surgery. We observed strong and persistent downregulation of immune and inflammatory pathways and significant remodeling of extracellular matrix components, especially at 5 years. Machine‑learning models accurately separated pre‑ and post‑surgery samples, and SHAP analysis confirmed that the most predictive genes aligned with biologically relevant pathways. Together, the analyses converged on a stable set of 50 genes that reflect long‑term reprogramming of adipose tissue. RYGB induces durable molecular changes in WAT characterized by reduced inflammation and structural remodeling that likely contribute to long‑term metabolic improvement. Integrating transcriptomics with explainable machine learning provides a powerful framework for identifying robust biomarkers and for gaining deeper insight into tissue adaptation after bariatric surgery.