An introductory review of parallel independent component analysis (p-ICA) and a guide to applying p-ICA to genetic data and imaging phenotypes to identify disease-associated biological pathways and systems in common complex disorders

Abstract

Complex inherited phenotypes, including those for many common medical and psychiatric diseases, are most likely underpinned by multiple genes contributing to interlocking molecular biological processes, along with environmental factors (Owen et al., 2010). Despite this, genotyping strategies for complex, inherited, disease-related phenotypes mostly employ univariate analyses, e.g., genome wide association. Such procedures most often identify isolated risk-related SNPs or loci, not the underlying biological pathways necessary to help guide the development of novel treatment approaches. This article focuses on the multivariate analysis strategy of parallel (i.e., simultaneous combination of SNP and neuroimage information) independent component analysis (p-ICA), which typically yields large clusters of functionally related SNPs statistically correlated with phenotype components, whose overall molecular biologic relevance is inferred subsequently using annotation software suites. Because this is a novel approach, whose details are relatively new to the field we summarize its underlying principles and address conceptual questions regarding interpretation of resulting data and provide practical illustrations of the method.

Keywords: common disease; common variant; genetic risk; imaging genetics; multivariate; parallel independent component analysis.

FIGURE 1

Univariate approaches are focused on single points of relation whereas multivariate approaches like parallel ICA (p-ICA) focus on links between patterns (e.g., weighted combinations of brain regions and weighed combinations of genetic variables.

FIGURE 2

The benefit of a joint analysis is we can capitalize on the joint distribution of (in this case) the imaging and genetic data, something that can provide a better ability to discriminate health and disease. When we have two data sets, each with numerous variables, we could compute huge numbers of cross-correlations (adjusting for requisite multiple comparisons). Here, p-ICA displays a definite advantage, providing both a means to identify relationships among two very large data sets, while simultaneously identifying the most relevant variables representing this information, (i.e., simultaneously performing data reduction).

FIGURE 3

Weighted combinations of brain regions are linked to weighted combinations of genetic variables which can then be tested for associations with variables of interest (e.g., disease status, symptoms). Components extracted by p-ICA are a linear weighted combination of all variables. Each variable’s weight indicates its contribution to the component, and helps to interpret it. For instance, the genetic component, perhaps formed from thousands of SNP markers, is mainly contributed to by top-weighted markers. The remainder, with much lower weights do not markedly affect the component loading.

FIGURE 4

Weighted combinations of brain regions are linked to weighted combinations of genetic variables which can then be tested for associations with variables of interest (e.g., disease, symptoms).