Integrated Multi-Omics Identification of Novel Diagnostic Biomarkers and Immunometabolic Therapeutic Targets in Osteoporosis via Machine Learning.

Abstract

Introduction: Osteoporosis (OP) is a multifactorial skeletal disorder marked by reduced bone mass and microstructural deterioration. We investigated potential diagnostic biomarkers and immunometabolic targets through integrated bioinformatics and machine learning analysis.

Methods: Batch-corrected GEO transcriptomic data were analyzed by WGCNA to identify OPassociated modules. An optimal diagnostic model was established using 12 machine learning algorithms (113 combinations). Immune infiltration and pathway activity were evaluated by CIBERSORT and GSEA, and hub genes were validated by qRT-PCR in plasma samples.

Results: The best random forest-based model had good predictive performance for both the training dataset and validation dataset (AUC > 0.80). Ten hub genes were finally screened out as candidate diagnostic biomarkers: POLA1, ARAP1, MAP7D1, TSPYL2, MRPL4, PPP1R14D, INTS1, SMARCB1, DLK1, and FPGT. Functional analyses showed enrichment in immunerelated and metabolic pathways, including PI3K-Akt signaling and pentose phosphate metabolism. These genes correlated with immune cell alterations, such as CD8⁷ T cells, Th17 cells, and macrophages. qRT-PCR validation confirmed their significant differential expression in OP samples.

Discussion: This study underscores immune-metabolic dysregulation at the core of OP pathobiology. Although heterogeneity existed in external validation cohorts, and sample sizes were small, our integrative analysis pipeline enhances the precision and clinical relevance of potential biomarkers.

Conclusion: The study concluded that this work identified ten possible diagnostic biomarkers, and provided a mechanism for understanding the interaction between immune metabolism and OP, which will help guide further research on precision diagnosis and treatment in OP.