Novel Gene Signature Classification Model for Predicting Severity of Skin Fibrosis

Scleroderma or systematic sclerosis (SSc) is a clinically heterogeneous disease with a complex phenotype. The disease is characterized by vascular dysfunction, tissue fibrosis, internal organ dysfunction, and immune dysfunction. Because it has no validated biomarkers or effective disease‐modifying therapies, scleroderma represents a major unmet medical need. Although a rare disease those afflicted with the systemic form of the disease suffer from severe vascular and other organ manifestations including pulmonary fibrosis. The leading cause of death is end‐organ dysfunction caused by fibrosis. An improved understanding of the molecular mechanisms involved in the initiation and perpetuation of fibrosis is mandatory to facilitate the rational design of new diagnostic and therapeutic strategies to detect earlier, arrest, and perhaps even reverse disease. We have also determined that chemokines, cytokines that help recruit immune cells to sites of infection and function in the migration of cells to aid in tissue repair, are key to ECM regulation. Diverse expressions of chemokine receptors on multiple dermal cellular populations are likely driven by inflammatory stimuli from excessive fibrosis. The contribution of chemokines to the human immune response is known, but its direct role on fibroblast‐induced ECM synthesis resulting in SSc disease severity is less clear. We three constructed classification models for making predictions on the severity of SSc, based on (1) gene expression profiles and (2) identifying marker gene sets that correlate with the classification. We utilized two existing genome‐wide microarray gene expression data from the Gene Expression Omnibus (GEO) database of SSc patients. Patients were divided into two groups based on skin score: low 0–17 (low‐mild) and high 18–51 (moderate‐sever modified Rodnan skin thickness score (mRSS), a predictor of disease outcome. In both the Group 1 and Group 2 dataset, random forest (RF) classifier combined with correlation‐based feature selection method (CFS) and naive bayes (NB) combined with CFS or support vector machine based recursive feature elimination method (SVM‐RFE) performed better than all other classification and feature selection methods. We then perform DAVID‐based GO enrichment analysis to evaluate if the chemokine pathways are enriched among differentially expressed genes. First, these data support the feasibility of accurately aligning the phenotype of each subset of patients and individual patient mRSS to generate gene signatures. Second, we uncovered differential expression of chemokine and ECM genes profiles that could potentially predict diseases severity