Integrative Machine Learning and Network Analysis of Skeletal Muscle Transcriptomes Identifies Candidate...

Area of Science:

• Genomics and Bioinformatics
• Endocrinology and Metabolism
• Precision Medicine

Background:

• Polycystic ovary syndrome (PCOS) is a common endocrine-metabolic disorder.
• Skeletal muscle insulin resistance in PCOS contributes significantly to cardiometabolic risk.
• Pioglitazone improves insulin sensitivity in PCOS, but its molecular mechanisms and biomarker potential are not fully understood.

Purpose of the Study:

• To reanalyze skeletal muscle gene expression data from pioglitazone-treated PCOS patients.
• To identify candidate biomarkers and regulatory hubs using machine learning and network approaches.
• To explore potential for precision therapy in PCOS.

Main Methods:

• Reprocessing of public microarray data (GSE8157) from PCOS and control skeletal muscle.
• Identification of differentially expressed genes (DEGs) and pathway analysis (Ingenuity Pathway Analysis).
• Application of four supervised machine learning algorithms with cross-validation for gene analysis.
• Construction of gene co-expression networks to identify hub genes.
• Simulated multi-omics framework integrating transcripts and clinical variables.

Main Results:

• 1459 DEGs were identified in PCOS skeletal muscle post-pioglitazone, involving immune, fibrotic, interferon, and epigenetic signaling.
• Machine learning models achieved excellent discrimination between PCOS and controls using a compact gene panel.
• Key regulatory nodes identified include ITK, WT1, BRD1-linked loci, and long non-coding RNAs.
• Simulated multi-omics signatures showed stabilized discovery performance.

Conclusions:

• Integrative machine learning and network analysis revealed candidate genes and regulatory hubs distinguishing PCOS.
• Findings suggest immunometabolic and epigenetic roles for pioglitazone in PCOS.
• ITK, WT1, BRD1-associated loci and network genes are nominated as promising biomarkers for validation in larger cohorts.