Block-based Bayesian epistasis association mapping with application to WTCCC type 1 diabetes data

Yu Zhang, Jing Zhang, Jun S. LiuThe Annals of Applied Statistics 2011 · DOI · arXiv

Abstract

Interactions among multiple genes across the genome contribute significantly to the risks of complex human diseases. While whole-genome single nucleotide polymorphism (SNP) data offer a path to understanding these interactions, the high correlation between nearby SNPs—known as linkage disequilibrium (LD)—and the sheer volume of possible interactions present massive computational hurdles. This paper introduces BEAM2, a novel Bayesian framework designed to simultaneously partition SNPs into LD-blocks and identify disease-associated variants. By modeling the genome's natural block structure, BEAM2 effectively filters out redundant associations and improves the detection of multi-locus interactions. Applied to the WTCCC Type 1 Diabetes (T1D) dataset, the method identified several known risk genes and revealed novel two-way associations within the MHC region.

Introduction

Identifying the genetic architecture of complex diseases requires moving beyond single-marker analysis. Most common diseases are polygenic, involving the interplay of many genetic variants, or epistasis. However, two primary challenges have hindered the discovery of these interactions: the combinatorial explosion of testing all possible SNP pairs and the confounding effects of Linkage Disequilibrium (LD). Standard methods often struggle to distinguish a causal variant from its neighbors that are inherited alongside it, leading to high false-positive rates.

To address these issues, the authors propose BEAM2, a block-based Bayesian association mapping tool. Unlike previous iterations that relied on simpler correlation models, BEAM2 treats the genomic block structure as a variable to be inferred, allowing for a more biologically grounded analysis of genetic risk.

The BEAM2 Framework

Simultaneous Inference and Block Partitioning

The core innovation of BEAM2 is its ability to simultaneously infer block boundaries and disease associations using Markov Chain Monte Carlo (MCMC) sampling. Instead of relying on pre-defined, "hard" block boundaries, the model provides posterior probabilities for block transitions. This "soft partitioning" reflects the inherent uncertainty in genomic structures and recombination hotspots.

Diplotype Modeling and SNP Classification

Rather than analyzing SNPs in isolation, BEAM2 evaluates "diplotypes"—the specific combinations of genotypes within an inferred block. This approach captures more information than single-marker tests and allows the model to classify every SNP into one of three functional groups:

  • Group 0: Variants unrelated to the disease.
  • Group 1: Variants with a marginal association (acting independently).
  • Group 2: Epistatic variants (acting in interaction with others).

By modeling blocks, the algorithm implicitly filters out "passenger" SNPs. If an entire block shows correlation with a disease, the Bayesian model prioritizes the most informative SNPs within that block, reducing the noise typically found in high-LD regions.

Experimental Validation

Simulation and Performance

The authors validated BEAM2 using simulated data generated via HAPGEN to mimic realistic human genetic structures. The results demonstrated that BEAM2 consistently outperformed the original BEAM and other block-based methods in detecting multi-locus associations, particularly as the frequency of disease-associated variants increased. Furthermore, when applied to the HLA region of 50 individuals, the model’s inferred block boundaries aligned closely with experimentally verified recombination hotspots.

Application to Type 1 Diabetes (WTCCC)

The method was applied to the Wellcome Trust Case Control Consortium (WTCCC) T1D dataset, comprising 2,000 cases and 3,004 controls. After filtering for quality, the analysis focused on approximately 42,470 SNPs. BEAM2 successfully recovered well-known T1D-associated genes, including PTPN22, CTLA4, MHC, and IL2RA.

More importantly, the model revealed complex interactions within the Major Histocompatibility Complex (MHC) region. By controlling for the haplotypes of the MHC class II region, the authors identified additional associations in MHC class I (HLA-A, HLA-B) and class III regions (BAT1). Significant interactions were also observed between the genes PRSS16 and ZNF184 (located in the extended MHC region) and MHC class II genes. These findings highlight the power of joint association mapping; many of these signals are undetectable by traditional single-SNP methods.

Discussion and Future Directions

While BEAM2 represents a significant step forward, it is computationally intensive. Analyzing a full modern GWAS (500k+ SNPs) currently requires pre-filtering or significant memory resources. The authors suggest that future iterations could benefit from parallel tempering to improve MCMC convergence and the integration of genotype imputation to handle missing data.

In conclusion, BEAM2 provides a robust statistical framework for navigating the complexities of the human genome. By treating LD as a structural feature rather than a nuisance, it offers a clearer view of the epistatic interactions that drive complex human diseases.

Summary updated

Abstract

Interactions among multiple genes across the genome may contribute to the risks of many complex human diseases. Whole-genome single nucleotide polymorphisms (SNPs) data collected for many thousands of SNP markers from thousands of individuals under the case-control design promise to shed light on our understanding of such interactions. However, nearby SNPs are highly correlated due to linkage disequilibrium (LD) and the number of possible interactions is too large for exhaustive evaluation. We propose a novel Bayesian method for simultaneously partitioning SNPs into LD-blocks and selecting SNPs within blocks that are associated with the disease, either individually or interactively with other SNPs. When applied to homogeneous population data, the method gives posterior probabilities for LD-block boundaries, which not only result in accurate block partitions of SNPs, but also provide measures of partition uncertainty. When applied to case-control data for association mapping, the method implicitly filters out SNP associations created merely by LD with disease loci within the same blocks. Simulation study showed that this approach is more powerful in detecting multi-locus associations than other methods we tested, including one of ours. When applied to the WTCCC type 1 diabetes data, the method identified many previously known T1D associated genes, including PTPN22, CTLA4, MHC, and IL2RA. The method also revealed some interesting two-way associations that are undetected by single SNP methods. Most of the significant associations are located within the MHC region. Our analysis showed that the MHC SNPs form long-distance joint associations over several known recombination hotspots. By controlling the haplotypes of the MHC class II region, we identified additional associations in both MHC class I (HLA-A, HLA-B) and class III regions (BAT1). We also observed significant interactions between genes PRSS16, ZNF184 in the extended MHC region and the MHC class II genes. The proposed method can be broadly applied to the classification problem with correlated discrete covariates.
BibTeX
@article{zhang2011block,
  author = {Yu Zhang and Jing Zhang and Jun S. Liu},
  title = {{Block-based Bayesian epistasis association mapping with application to WTCCC type 1 diabetes data}},
  journal = {The Annals of Applied Statistics},
  year = {2011},
  doi = {10.1214/11-aoas469},
}
Figure