Introduces PMGCN, a percolation-based method to find parsimonious multi-disease gene co-expression biomarkers from bulk transcriptomes, validated mainly on ulcerative colitis and other inflammatory diseases but not producing PD-specific biomarkers.

The approach could help prioritize inflammation-linked biomarkers that might connect periodontitis and Parkinson's for downstream validation, but its immediate therapeutic relevance to PD is limited by lack of direct PD results or mechanistic/targetable findings.

BACKGROUND: Chronic inflammatory diseases, such as ulcerative colitis (UC), Crohn's disease (CD), Alzheimer's disease (AD) and Parkinson's disease (PD) are clinically related to periodontitis. However, the computation of omic biomarkers regarding these diseases has not leveraged this association. METHODS: We developed PMGCN, a computational framework that employs optimal percolations on multi-disease gene co-expression networks derived from bulk transcriptomic gene expression profiles to identify a parsimonious set of key nodes as candidate omic biomarkers. RESULTS: Evaluation of PMGCN on independent clinical studies of four chronic inflammatory diseases demonstrates improved predictive performance evaluated via cross validation with bootstrapping compared to commonly used univariate differentially expressed genes. Specifically for UC, three key gene biomarkers (CXCL5, FOSB, PTGR1) are identified by PMGCN, and public single-cell RNA-seq datasets confirm that the mainly altered inflammation signaling pathways in three cell clusters are connected to UC and periodontitis progression. CONCLUSIONS: PMGCN proposes a computational biomarker identification approach leveraging multi-disease association, the discovered gene biomarkers demonstrate improved prediction of chronic inflammatory diseases and provide novel insights into disease progression.