Identification of schizophrenia-associated gene polymorphisms using hybrid filtering feature selection with structural information

Abstract

Schizophrenia is a complex and severe neurological disorder that affects lots of people worldwide. Despite its strong evidence of heritability revealed by lots of genetic studies, research for locating of schizophrenia associated genes remains frustrating as numerous efforts had failed to identify biomarkers that could strongly impact the diagnosis and prognosis of schizophrenia. The major challenge lies in the weak discrimination of single gene marker and the enormous number of gene variants that exist in human genome. In this paper we propose a hybrid feature selection method that utilizes the biological structural information of the gene variants to tackle this problem. A set of statistical techniques are developed to encourage the clustering of multiple informative SNP variants on the same gene, which boost the probability of finding biologically meaningful features and suppresses false discoveries. As a result, the proposed method achieves significantly better performance on a published schizophrenia human genome data set compared with previous studies, with an area-under-ROC-curve of 65% and an odd ratio of 2.82 (95%CI: 1.80 – 4.40). 36 gene markers are discovered to be associated with the onset of schizophrenia with many of which verified directly or indirectly by previous literature. The method proposed in this paper can be also adopted for efficient control of false discoveries in finding biomarkers from genomic data.