Network and data integration for biomarker signature discovery via network smoothed T-statistics

Predictive, stable and interpretable gene signatures are generally seen as an important step towards a better personalized medicine. During the last decade various methods have been proposed for that purpose. However, one important obstacle for making gene signatures a standard tool in clinics is the typical low reproducibility of signatures combined with the difficulty to achieve a clear biological interpretation. For that purpose in the last years there has been a growing interest in approaches that try to integrate information from molecular interaction networks. We here propose a technique that integrates network information as well as different kinds of experimental data (here exemplified by mRNA and miRNA expression) into one classifier. This is done by smoothing t-statistics of individual genes or miRNAs over the structure of a combined protein-protein interaction (PPI) and miRNA-target gene network. A permutation test is conducted to select features in a highly consistent manner, and subsequently a Support Vector Machine (SVM) classifier is trained. Compared to several other competing methods our algorithm reveals an overall better prediction performance for early versus late disease relapse and a higher signature stability. Moreover, obtained gene lists can be clearly associated to biological knowledge, such as known disease genes and KEGG pathways. We demonstrate that our data integration strategy can improve classification performance compared to using a single data source only. Our method, called stSVM, is available in R-package netClass on CRAN (http://cran.r-project.org).