Beyond structural equation modeling: model properties and effect size from a Bayesian viewpoint. An example of complex phenotype-genotype associations in depression.

Despite the rapid evolution of measurement technologies in biomedicine and genetics, most of the recent studies aiming to explore the genetic background of multifactorial diseases were only moderately successful. One of the causes of this phenomenon is that the bottleneck of genetic research is no longer the measurement process related to various laboratory technologies, but rather the analysis and interpretation of results. The commonly applied univariate methods are inadequate for exploring complex dependency patterns of multifactorial diseases which includes nearly all common diseases, such as depression, hypertension, and asthma. A comprehensive investigation requires multivariate modeling methods that enable the analysis of interactions between factors, and allow a more detailed interpretation of studies measuring complex phenotype descriptors. In this paper we discuss various aspects of multivariate modeling through a case study analyzing the effect of the single nucleotide polymorphism rs6295 in the HTR1A gene on depression and impulsivity. We overview basic concepts related to multivariate modeling and compare the properties of two investigated modeling techniques: Structural Equation Modeling and Bayesian network based learning algorithms. The resulting models demonstrate the advantages of the Bayesian approach in terms of model properties and effect size as it allows coherent handling of the weakly significant effect of rs6295. Results also confirm the mediating role of impulsivity between the SNP rs6295 of HTR1A and depression.