Email updates

Keep up to date with the latest news and content from BMC Genetics and BioMed Central.

Open Access Highly Accessed Research article

Handling linkage disequilibrium in qualitative trait linkage analysis using dense SNPs: a two-step strategy

Kelly Cho2* and Josée Dupuis1

Author Affiliations

1 Department of Biostatistics, Boston University School of Public Health, Boston, MA, USA

2 Current address: Departments of Genetics and Biostatistics, Yale University Schools of Medicine and Public Health, PO BOX 208034, 300 George Street Suite 523, New Haven, CT 06520-8034, USA

For all author emails, please log on.

BMC Genetics 2009, 10:44  doi:10.1186/1471-2156-10-44

Published: 10 August 2009

Abstract

Background

In affected sibling pair linkage analysis, the presence of linkage disequilibrium (LD) has been shown to lead to overestimation of the number of alleles shared identity-by-descent (IBD) among sibling pairs when parents are ungenotyped. This inflation results in spurious evidence for linkage even when the markers and the disease locus are not linked. In our study, we first theoretically evaluate how inflation in IBD probabilities leads to overestimation of a nonparametric linkage (NPL) statistic under the assumption of linkage equilibrium. Next, we propose a two-step processing strategy in order to systematically evaluate approaches to handle LD. Based on the observed inflation of expected logarithm of the odds ratio (LOD) from our theoretical exploration, we implemented our proposed two-step processing strategy. Step 1 involves three techniques to filter a dense set of markers. In step 2, we use the selected subset of markers from step 1 and apply four different methods of handling LD among dense markers: 1) marker thinning (MT); 2) recursive elimination; 3) SNPLINK; and 4) LD modeling approach in MERLIN. We evaluate relative performance of each method through simulation.

Results

We observed LOD score inflation only when the parents were ungenotyped. For a given number of markers, all approaches evaluated for each type of LD threshold performed similarly; however, RE approach was the only one that eliminated the LOD score bias. Our simulation results indicate a reduction of approximately 75% to complete elimination of the LOD score inflation while maintaining the information content (IC) when setting a tolerable squared correlation coefficient LD threshold (r2) above 0.3 for or 2 SNPs per cM using MT.

Conclusion

We have established a theoretical basis of how inflated IBD information among dense markers overestimates a NPL statistic. The two-step processing strategy serves as a useful framework to systematically evaluate relative performance of different methods to handle LD.