Gene set analysis: A step-by-step guide

Abstract

To maximize the potential of genome-wide association studies, many researchers are performing secondary analyses to identify sets of genes jointly associated with the trait of interest. Although methods for gene-set analyses (GSA), also called pathway analyses, have been around for more than a decade, the field is still evolving. There are numerous algorithms available for testing the cumulative effect of multiple SNPs, yet no real consensus in the field about the best way to perform a GSA. This paper provides an overview of the factors that can affect the results of a GSA, the lessons learned from past studies, and suggestions for how to make analysis choices that are most appropriate for different types of data. © 2015 Wiley Periodicals, Inc.

Keywords: complex traits; gene set analysis; genome-wide association studies; polygenic effects.

FIG. 1

A gene set analysis workflow, including the possible choices that must be made at each step of the analysis. The data type requirements for the various statistical methods are indicated by color. For instance, regression-based methods and permutation procedures that randomize samples require genotypes as inputs. On the other hand, over-representation methods utilize summary statistics. Some methods (multicolored) are not restricted to one type of input data.

FIG. 2

Gene sets related to glucocorticoid receptor processes. Gene sets from NCI’s Pathway Interaction Database, the proprietary Metacore database, and the Gene Ontology database were overlaid onto an interaction network from the STRING protein-protein interaction database (only high confidence interactions are shown, STRING combined score ≥0.9). Colored genes are unique to a particular database, while gray genes are shared between two or more databases (only NR3C1 is common to all four databases). There are clear differences in membership between gene sets from different data sources. These differences may be due to an attempt to model distinct processes, but are also indicative of incomplete annotation. For instance, some genes are unique to a single database even when there is evidence of interaction with multiple genes from another database (e.g., ARID1A is not part of the NCI-PID gene set, but is connected to four of its member genes).

FIG. 3

Visualizations of gene set analysis results. Depending on the source of the gene sets, gene interaction information can be obtained from pathway maps or PPI databases. A: A signaling pathway map with genes colored to show gene-level association measures. Dark blue indicates a weak association and dark red indicates a strong association. B: A gene set overlaid onto a PPI network to show known interactions between genes. Here the strength of gene-level associations is indicated by node size.