Genetic Biomarker Selection for Obsessive-Compulsive Disorder by Sparse Representation Based Variable Selection Method

Background: Obsessive-compulsive disorder (OCD) is a debilitating neuropsychiatric condition estimated to afflict 1-3 % of the world population. Dozens of OCD candidate genes have been reported by an increased number of articles. Nevertheless, each patient/patient group may demonstrate unique etiologic characteristics that need personalized treatment. Method: We integrated a sparse representation based variable selection (SRVS) approach with an OCD-gene ResNet relation data analysis to select top genes for a specific group of 118 subjects, including 16 OCD cases and 102 healthy controls. The gene expression profile were acquired from the dorsolateral prefrontal cortex (DLPFC) of postmortem tissue of these subjects. A 77 OCD candidate genes were acquired from ResNet relation data analysis. Pathway enrichment analysis (PEA), sub-network enrichment analysis (SNEA) and gene-gene Interaction analysis (GGI) were conducted to study the functional profile of the top genes selected by SRVS, and compared with previous reported genetic markers. Results: A significantly high classification accuracy (CR) of 79.66 % was acquired (permutation p-value = 0.0046) using the top 9 genes selected by SRVS, including HOXB8, HTR2C, CRHR2, GRIK3, HGF, OXT, TPH2, DRD2 and ADRA1A. These genes were enriched within multiple pathways and sub-networks that were previously implicated with OCD. In contrast, using the same number of most frequently reported, a CR of only 65.5 % is achieved. Moreover, GGI results showed that these genes demonstrated a strong functional correlation with the frequently reported OCD genes. Conclusion: Our study suggests that SRVS is an effective method for data driven variable selection for OCD, and that the genes that were frequently reported to associate with OCD might not be the best biomarkers for a specific OCD patient/ patient group.