Gene coexpression network and machine learning in personalized psychiatry

Abstract Despite the success of collaborative international efforts to identify genetic variants involved in psychiatric disorders, the biological underpinnings of complex psychiatric traits remain elusive. Increasing evidence suggests that psychiatric disorders are the result of complex interactions among genomic variations, epigenetic modifications, and other regulatory mechanisms involved in gene expression. Therefore, the transcriptome, representing a nexus of genetic and environmental interactions, can be seen as an essential biological layer of information for studying molecular dysregulations in mental disorders. Transcriptomics can be used for diagnostic purposes to differentiate disease from healthy controls, differentiate disease stages, and identify subgroups of patients exhibiting different biological signatures within diagnosis. It also allows us to measure the influence of drugs on the transcriptome, which can be useful in gaining insights on molecular mechanisms of a drug’s action, and in predicting treatment response. Using coexpression network analysis-based methods, disease-relevant clusters of coregulated genes can be identified and further integrated with genetic and epigenetic data for a comprehensive investigation of biological underpinnings of mental illness. In this chapter, we provide an essential guide to the coexpression network approach as an important statistical tool that can enrich the understanding of disrupted molecular processes in psychiatric disorders. Furthermore, given the complex system structure inherent in psychiatric disorders, statistical learning frameworks that can translate these findings into actionable clinical insights are required. A number of emerging methodologies that address this problem are explored, including weighted gene coexpression network analysis, differential coexpression analysis, biclustering, and regularized machine learning.