Source-linked AI summary
BOOST: A fast approach to detecting gene-gene interactions in genome-wide case-control studies
Xiang Wan, Can Yang, Qiang Yang, Hong Xue, Xiaodan Fan, Nelson L. S. Tang, Weichuan Yu
TL;DR
Genome-wide gene-gene interaction detection is computationally and methodologically challenging. BOOST uses Boolean-operation screening and statistical testing to examine all SNP pairs quickly, analyzing seven WTCCC disease datasets in under 60 hours per dataset and revealing distinct T1D and RA interaction patterns plus many T1D MHC interactions.
Problem
Identifying gene-gene interactions in genome-wide case-control studies remains computationally and methodologically challenging, despite their importance for understanding complex disease genetics.
Method
BOOST combines Boolean genotype representation with a two-stage procedure that screens all pairwise interactions using an upper-bound statistic, then tests retained pairs with a likelihood-ratio χ2 test.
Results
BOOST analyzed seven WTCCC disease datasets, evaluating roughly 360,000 SNP pairs in under 60 hours per dataset and identifying many T1D MHC interactions with patterns distinct from RA.
Takeaways & Limitations
BOOST provides a computationally and statistically useful tool for large-scale interaction mapping and can reveal interaction patterns and loci not captured by single-locus scans.
Abstract
from arXiv · showhide
Gene-gene interactions have long been recognized to be fundamentally important to understand genetic causes of complex disease traits. At present, identifying gene-gene interactions from genome-wide case-control studies is computationally and methodologically challenging. In this paper, we introduce a simple but powerful method, named `BOolean Operation based Screening and Testing'(BOOST). To discover unknown gene-gene interactions that underlie complex diseases, BOOST allows examining all pairwise interactions in genome-wide case-control studies in a remarkably fast manner. We have carried out interaction analyses on seven data sets from the Wellcome Trust Case Control Consortium (WTCCC). Each analysis took less than 60 hours on a standard 3.0 GHz desktop with 4G memory running Windows XP system. The interaction patterns identified from the type 1 diabetes data set display significant difference from those identified from the rheumatoid arthritis data set, while both data sets share a very similar hit region in the WTCCC report. BOOST has also identified many undiscovered interactions between genes in the major histocompatibility complex (MHC) region in the type 1 diabetes data set. In the coming era of large-scale interaction mapping in genome-wide case-control studies, our method can serve as a computationally and statistically useful tool.
Results
BOOST evaluated all pairwise interactions among roughly 360,000 SNPs across seven WTCCC disease datasets in under 60 hours per dataset. It revealed contrasting MHC interaction patterns between type 1 diabetes and rheumatoid arthritis, including interactions lacking significant single-locus effects.
- Results: Under 60 hours per dataset, BOOST evaluated all pairs of roughly 360,000 SNPs across seven WTCCC disease datasets on a standard desktop.The analyses used 14,000 cases in total and 3,000 shared controls.
- Results: BOOST reported 4,499 interactions in type 1 diabetes, with 4,489 (99.8%) in the MHC region, versus 350 interactions in rheumatoid arthritis, with 280 (80.0%) there.Both diseases were strongly associated with the MHC region in single-locus mapping, but their interaction results differed.
- Results: For the other six diseases, BOOST found no nontrivial interactions except one interacting SNP pair in Crohn’s disease.The study contrasts this finding with earlier methods whose computational scalability or interaction-detection scope was limited.
- Results: Type 1 diabetes showed strong interactions within and across MHC classes I, III, and II, whereas rheumatoid arthritis interactions mostly involved closely placed loci in class II.The contrast may explain different aetiologies not revealed by single-locus association mapping.
- Results: Among 789 selected type 1 diabetes interaction pairs, 91 had non-significant loci under the single-locus scan.These interactions included patterns between MHC class I and class II regions, including HLA-B and a block across HLA-DQA2 and HLA-DQB2.
- Results: BOOST’s Boolean genotype representation improves space and CPU efficiency by enabling fast bitwise operations for contingency-table construction.The method combines this representation with screening and testing stages to handle genome-wide interaction searches.
METHODS
BOOST uses Boolean genotype encoding and a two-stage screening-and-testing strategy to evaluate genome-wide SNP interactions efficiently. It approximates interaction statistics during screening, then applies likelihood-ratio testing to selected pairs.
- Data representation: Boolean genotype representation uses binary values to improve space and CPU efficiency during multi-locus analysis.The original genotype matrix stores values in {1, 2, 3}; BOOST instead uses Boolean operations for faster logic.
- Interaction measurement: Directly evaluating likelihood-ratio interaction statistics is computationally unaffordable because genome-wide studies contain hundreds of billions of SNP pairs.The interaction statistic compares full and main-effect models and is asymptotically χ2 distributed with df = 4.
- Interaction measurement: Log-linear models provide an equivalent formulation for measuring interaction effects, with ˆLS −ˆLH linked to Kullback-Leibler divergence.MH corresponds to the main-effect model, while MS corresponds to the full model.
- Screening: Kirkwood Superposition Approximation provides a non-iterative approximation for the otherwise computationally intensive estimation of ˆp.The approximation yields ˆpK and its associated log-likelihood ˆLKSA.
- Screening: BOOST screens every SNP pair with KSA and Boolean contingency-table operations, retaining pairs whose statistic exceeds τ = 30.Because 2(ˆLS −ˆLH) ≤ 2(ˆLS −ˆLKSA), pairs below the screening threshold are excluded from testing.
- Testing: In Stage 2, selected pairs are tested with the likelihood-ratio statistic 2(ˆLS −ˆLH), followed by a χ2 test with df = 4.The testing stage fits the log-linear models MH and MS to measure interaction effects.