Directed Network Analysis of Genome-Wide Data for Post Traumatic Stress Disorder and Fibromyalgia Diagnosis

Abstract

Network-based computational analysis of genome-wide data has been used to aid prediction of disease status with success in the past. Machine learning classifiers are usually applied to datasets of gene expression and/or methylation in order to classify a patient as diseased or healthy. Nevertheless, the large number of genes usually contained in these datasets makes it difficult to effectively identify genes with high relevance to the disease condition. The integration of information on interactions between genes, proteins, and other biochemical entities from databases of known biological pathways has been used to improve the analysis of abnormalities in gene expression and methylation with encouraging results [1,2,3]. Chuang et al. explored the effects that using combined subsets of measurements in known biological pathways has in identifying biological markers for disease classification [4]. The idea relies on the concept that biologically linked components of a network tend to have similar measurements(expression or methylation) when they are close to one another. In this project, the Chuang et al.algorithm is extended in order to exploit the effect of directionality on networks on identifying disease status [4]. This extended algorithm is applied to datasets on two diseases: Fibromyalgia and PostTraumatic Stress Disorder (PTSD). The results of the algorithm are then used as features in nine different classification algorithms. The results show that it is possible to achieve an accuracy rate similar, and in some cases better, than using individual methylation and expression measurements of genes as features with much less computational power. Furthermore, the use of this technique yields insight into the relationship between these often comorbid disorders.