Statistical and Machine Learning Approaches for Network Analysis

9.4 Summary and Conclusions

In this study, we proposed a network-based information synergy approach to identify candidate genes involved in AD. Results obtained from simulation data and AD microarray data suggested that, information synergy, particularly positive information synergy, could identify gene pairs with specific joint expression patterns that would otherwise be overlooked by differential and correlation analyses. Meanwhile, some of the hub genes in the PPI subnetworks, consisting of positive information synergy interactions, show biological relevance to the pathogenesis of AD, suggesting that the networks obtained through information synergy could potentially identify genes involved in diseases. Moreover, the information theory used in synergy analysis allows one to capture gene pairs with different types of relationship (either linear or nonlinear), as long as the gene pairs provide additional information on the phenotype. This advantage renders information synergy particularly attractive for capturing gene pairs in complex scenarios, where the interactions between genes are not linear. Taken together, information synergy is a promising complementary approach to network-based studies.

The concept of information synergy has been applied to identify gene pairs predictive of phenotypes based upon microarray data. The synergistic gene pairs inferred from microarray data may not necessarily interact with each other physically. This makes it difficult to interpret the biological ...