Schizophrenia prediction using integrated imaginggenomic networks

In order to increase the diagnosis accuracy of schizophrenia (SCZ) disease, it is essential to comprehensively employ complementary information from multiple types of data. It is well known that a network is a general method for analyzing relationships between patients, with its nodes representing patients and its edges showing relationships between them. In this study, we constructed a fused network using three types of data including genetic, epigenetic and neuroimaging data from a study of schizophrenia patients. We developed a network-based prediction approach taking advantage of the whole network of patients rather than just individual clusters in the network. The majority neighborhood of a node in the network was exploited as a feature for discriminating SCZ from healthy controls. Compared with other 9 graph-based label prediction methods, our network-fusion based label prediction method shows the best performance according to the prediction accuracy. The prediction power of our proposed method was also tested under different parameters settings and an optimal parameter was found for achieving the best performance. The method is also computationally efficient and can be extended to identify other clinical outcomes.