A fast algorithm for detecting gene–gene interactions in genome-wide association studies

Jiahan Li, Wei Zhong, Runze Li, Rongling WuThe Annals of Applied Statistics 2014 · DOI · arXiv

Abstract

With the advent of high-throughput genotyping, genome-wide association studies (GWAS) have become essential for identifying genetic drivers of complex traits. However, traditional methods often fail to account for gene-gene interactions (epistasis) due to the overwhelming dimensionality of SNP-SNP combinations. This paper introduces a novel statistical framework featuring a two-stage sure independence screening (TS-SIS) procedure and a rates adjusted thresholding estimation (RATE) approach. By efficiently reducing the search space while allowing for weak heredity, the framework identifies significant epistatic interactions that contribute to phenotypic variation. Application to the Framingham Heart Study data reveals 23 active SNPs and 24 epistatic interactions associated with Body Mass Index (BMI), demonstrating the framework's ability to resolve the complexities of genetic control.

Introduction

In modern genetics, the search for single-nucleotide polymorphisms (SNPs) associated with complex diseases often overlooks the collaborative nature of biology. While single-SNP analysis has been the standard, it is widely recognized that epistasis—the interaction between different genes—plays a pivotal role in genetic variation.

The primary barrier to studying epistasis is the "curse of dimensionality." In a typical GWAS with 500,000 SNPs, there are approximately 125 billion possible pairwise interactions. Most existing statistical tools rely on "strong heredity" assumptions, which only test interactions between two SNPs if both are individually significant. This paper addresses the more challenging problem of "weak heredity," where an interaction may be significant even if one or both of the individual SNPs show little marginal effect.

Methodology

The proposed framework utilizes a three-step pipeline designed to maintain computational efficiency without sacrificing discovery power:

1. Two-Stage Sure Independence Screening (TS-SIS)

To avoid the impossible task of testing every possible pair, the TS-SIS procedure operates in two stages:

  • Stage 1: All individual SNPs are screened to identify a pool of "candidate roots" that demonstrate at least a marginal correlation with the phenotype.
  • Stage 2: The algorithm tests interactions between these candidate roots and the rest of the genome. This targeted approach drastically reduces the search space while remaining flexible enough to capture weak heredity interactions.

2. Rates Adjusted Thresholding Estimation (RATE)

Determining where to "cut off" the screening process is a critical challenge. The authors propose RATE, a data-driven thresholding technique. By utilizing bootstrapping and auxiliary variables (synthetic "noise" data), RATE calibrates a threshold that controls the false positive rate at a specific confidence level, ensuring that the reduced model remains manageable yet inclusive of true signals.

3. Penalized Regression

Once the billions of potential interactions are narrowed down to a few hundred candidates, the framework applies regularization methods such as LASSO or SCAD (Smoothly Clipped Absolute Deviation). These "shrinkage" techniques perform final variable selection to identify the most impactful genetic drivers.

Key Insights and Results

Resolving "Missing Heritability"

A recurring mystery in genomics is why single-SNP associations explain so little of a trait's variation. This study suggests that epistasis is the missing piece of the puzzle. In the analysis of BMI variation within the Framingham Heart Study (FHS) dataset, interactions explained 23% of the phenotypic variation, compared to only 16.1% explained by main effects.

Performance in the Framingham Heart Study

The framework was applied to a dataset of 349,985 SNPs from 977 individuals. The model identified:

  • 23 active SNPs and 24 active epistatic interactions.
  • A classification accuracy of 82.77% when predicting obesity (BMI > 30).
  • Biological relevance in the findings, such as SNP ss66142093 near the ANAPC13 gene, which is linked to cell cycle regulation and bone development—pathways previously associated with obesity and human stature.

Computational Efficiency

Unlike traditional methods that might require supercomputing clusters to process billions of interactions, the TS-SIS approach is computationally lean. It allows these complex calculations to be performed on standard hardware in a fraction of the time, making large-scale epistatic analysis accessible to more researchers.

Conclusion

The TS-SIS and RATE framework represents a significant step forward in GWAS methodology. By moving beyond the limitations of strong heredity and single-SNP analysis, this approach provides a scalable solution for uncovering the complex genetic architecture of human traits. Future work will focus on extending this framework to binary traits (case-control studies) and exploring higher-order interactions beyond pairwise epistasis.

Summary updated

Abstract

With the recent advent of high-throughput genotyping techniques, genetic data for genome-wide association studies (GWAS) have become increasingly available, which entails the development of efficient and effective statistical approaches. Although many such approaches have been developed and used to identify single-nucleotide polymorphisms (SNPs) that are associated with complex traits or diseases, few are able to detect gene-gene interactions among different SNPs. Genetic interactions, also known as epistasis, have been recognized to play a pivotal role in contributing to the genetic variation of phenotypic traits. However, because of an extremely large number of SNP-SNP combinations in GWAS, the model dimensionality can quickly become so overwhelming that no prevailing variable selection methods are capable of handling this problem. In this paper, we present a statistical framework for characterizing main genetic effects and epistatic interactions in a GWAS study. Specifically, we first propose a two-stage sure independence screening (TS-SIS) procedure and generate a pool of candidate SNPs and interactions, which serve as predictors to explain and predict the phenotypes of a complex trait. We also propose a rates adjusted thresholding estimation (RATE) approach to determine the size of the reduced model selected by an independence screening. Regularization regression methods, such as LASSO or SCAD, are then applied to further identify important genetic effects. Simulation studies show that the TS-SIS procedure is computationally efficient and has an outstanding finite sample performance in selecting potential SNPs as well as gene-gene interactions. We apply the proposed framework to analyze an ultrahigh-dimensional GWAS data set from the Framingham Heart Study, and select 23 active SNPs and 24 active epistatic interactions for the body mass index variation. It shows the capability of our procedure to resolve the complexity of genetic control.
BibTeX
@article{li2014fast,
  author = {Jiahan Li and Wei Zhong and Runze Li and Rongling Wu},
  title = {{A fast algorithm for detecting gene–gene interactions in genome-wide association studies}},
  journal = {The Annals of Applied Statistics},
  year = {2014},
  doi = {10.1214/14-aoas771},
}
Figure