A Novel Bioinformatics Strategy to Explore Key Biomarkers in IgA Nephropathy Using Network Biomarkers and Machine Learning

Background: Serum metabolites have been established as pivotal biomarkers for a multitude of diseases. This study explored serum metabolic differences between IgAN patients and healthy controls, aiming to identify potential biomarkers for IgAN. Methods: Serum from 31 healthy individuals and 65 IgAN patients was analyzed using ultra-performance liquid chromatography-mass spectrometry. Machine learning algorithms combined with traditional molecular/ network biomarkers differentiated IgAN from controls, identifying key metabolites and interaction networks. All subjects were followed for >3 years to assess their prognostic value, with mechanisms explored by Mendelian Randomization(MR), network pharmacology, and molecular docking. Results: Volcano plot and PLS-DA analyses showed distinct serum metabolite profiles in IgAN patients compared to healthy controls, correlating with kidney damage. A machine learning model utilizing network biomarkers also surpassed traditional biomarkers in performance. SHAP analysis recognized Dehydroepiandrosterone sulfate (DHEA-S) as a crucial differentiator, and networks differed significantly between IgAN cases and controls, supporting its strong linkage to IgAN. MR analysis established a causal link between reduced DHEA-S levels and IgAN onset. Long-term follow-up revealed higher DHEA-S levels correlated with better renal outcomes. Network pharmacology identified 42 potential targets and a protein-protein interaction network pinpointed 6 central targets; molecular docking showed stable binding of DHEA-S with 4 proteins, suggesting its involvement in signaling pathways and mechanisms in IgAN. Conclusion: The network structure of serum metabolites in IgAN was significantly different from that of healthy controls, with changes of DHEA-S closely related to the progression of IgAN.