Volume 6 Supplement 1
Genetic Analysis Workshop 14: Microsatellite and singlenucleotide polymorphism
HasemanElston weighted by marker informativity
 Daniel Franke^{1},
 André Kleensang^{1},
 Robert C Elston^{2} and
 Andreas Ziegler^{1}Email author
DOI: 10.1186/147121566S1S50
© Franke et al; licensee BioMed Central Ltd 2005
Published: 30 December 2005
Abstract
In the HasemanElston approach the squared phenotypic difference is regressed on the proportion of alleles shared identical by descent (IBD) to map a quantitative trait to a genetic marker. In applications the IBD distribution is estimated and usually cannot be determined uniquely owing to incomplete marker information. At Genetic Analysis Workshop (GAW) 13, Jacobs et al. [BMC Genet 2003, 4(Suppl 1):S82] proposed to improve the power of the HasemanElston algorithm by weighting for information available from marker genotypes. The authors did not show, however, the validity of the employed asymptotic distribution. In this paper, we use the simulated data provided for GAW 14 and show that weighting HasemanElston by marker information results in increased type I error rates. Specifically, we demonstrate that the number of significant findings throughout the chromosome is significantly increased with weighting schemes. Furthermore, we show that the classical HasemanElston method keeps its nominal significance level when applied to the same data. We therefore recommend to use HasemanElston with marker informativity weights only in conjunction with empirical pvalues. Whether this approach in fact yields an increase in power needs to be investigated further.
Background
The Haseman and Elston (HE) method [1] is the best known approach to map quantitative traits by linkage analyses. It has been extended in many different ways to increase statistical power. For example, Amos et al. [2] introduced a generalized least squares approach and weighted the squared phenotypic differences inversely proportional to their respective variances. Sham and Purcell [3] also proposed to weight phenotypes: they combined the squared trait difference and the squared mean centered sum linearly, the weights given to these two components depending on the overall trait correlation.
A different approach has been followed by Jacobs et al. [4] at the Genetic Analysis Workshop (GAW) 13. They proposed to weight families for HE linkage analyses according to marker informativity (as measured by the difference between the allele sharing at the marker and the allele sharing for a noninformative marker) and applied their method to genome scan data for blood pressure. The anticipated gain in power is substantial. For example, the pvalue dropped from approximately 10^{2} to 10^{4} on chromosome 5 and from 10^{3} to 10^{5} on chromosome 12 for the mean systolic blood pressure. However, Jacobs et al. [4] did not show the validity of their approach, i.e., the asymptotic normality of the proposed test statistic.
Here, we pick up their work. We use the same distance metric as Jacobs et al. as well as a different simplexbased weighting scheme and show that the HE regression with weights according to this measure of marker informativity suffers from inflated type I error rates. We illustrate the effect of the weighted HE regression using data from chromosome 4 of the Aipotu population as available for GAW 14. For the analyses, we required and utilized the correct answers for the simulated data.
Simplex weighting scheme
In the simplex weighting scheme, families are weighted according to their degree of informativity for linkage. The degree of informativity is determined by the Euclidian distance d between the current available identitybydescent (IBD) marker information and the IBD information for uninformative markers.
If a genetic marker is noninformative, the IBD distribution equals (1/4,1/2,1/4). We therefore define simplex weights w as the Euclidian distance d between a marker with IBD distribution f = (f_{0}, f_{1}, f_{2}) and a noninformative marker by
Jacobs et al. [4] proposed slightly different weights and determined marker informativity by a distance metric D defined as
Other measures of informativity might be preferable but are beyond the scope of this paper.
Classical HE and weighted HE regression
For simplicity, we consider a sample of n independent sib pairs. Then, the classical HE regression ignores a possible dominance effect and fits the linear model
Δ^{ 2 }= X_{ γ }+ ε, where
and the identity matrix Ω. The squared phenotypic difference of the i^{th} sib pair is represented by , and π_{ i }denotes the proportion of alleles IBD (π_{ i }= f_{i 2}+ f_{i 1}/ 2) of the i^{th} sib pair. In practice, a ttest statistic is employed to test the null hypothesis of no linkage. Under H_{0} the test statistic
is asymptotically distributed as t_{n1}, where n  1 denotes the degrees of freedom. The parameter estimators for β and are given by
As soon as weights are introduced, Ω no longer represents the identity matrix. Specifically, we aim to use simplex weights as described above, so that Ω becomes a diagonal matrix with elements w_{ i }(f). Since the IBD distribution is estimated from the available marker data, replaces the true matrix Ω. Test statistics and estimates are obtained from equations (3) and (4) where Ω is replaced by its estimate .
Methods
We illustrate the effect of the weighted HE regression by using data from chromosome 4 of Aipotu. The ethnic group as well as the chromosome was randomly chosen. No positive finding in the microsatellite scan can be expected for chromosome 4.
We employed a development branch of SIBPAL and GENIBD from S.A.G.E. [5] to compute three different asymptotic HE regression models for each of the 100 replicates: the classical HE method as well as HE weighted by weights according to Equations (1) and (2), respectively. GENIBD was utilized to estimate multipoint IBD distributions; estimates were obtained at the marker positions.
We wished to investigate the validity of the asymptotic distribution of the weighted HE methods. This can be achieved by testing whether the number of significant findings across a chromosome is significantly increased. For sake of simplicity, we used the 5% test level for all further investigations. For each genetic marker position, each of the 100 replicates and each weighting scheme, p < 0.05 was tested. Generalized estimating equations with an autoregressive (AR(1)) working correlation structure, using the replicate as class level indicator, was used for each weighting scheme to investigate whether the proportion of significant findings across the chromosome exceeded 5%. This model adequately adjusts for intermarker correlations on a chromosome. The estimated proportion of significant findings across the whole chromosome is reported with its corresponding asymptotic 95% confidence interval (CI).
Furthermore, we want to show that the number of significant findings is greater using the weighted HE methods compared with the classical HE approach. To this end, a Wilcoxon signed rank test was employed. Specifically, we counted the number of microsatellite positions where p < 0.05 for both weighted and the classical HE regression across the whole chromosome. The Wilcoxon signed rank tested was computed across the 100 independent chromosomes. If the weighted HE methods was too liberal, the number of significant findings would be significantly increased compared with the classical HE method.
Results
Number of onesided pvalues smaller than 5% in the 100 replicates of chromosome 4 from the simulated dataset of Aipotu.
Weights  

Locus  HE method  Simplex weights (Eq. 1)  Jacobs et al. [4] method 
D04S0128  8  9  9 
D04S0129  5  6  5 
D04S0130  5  6  6 
D04S0131  5  5  5 
D04S0132  8  9  9 
D04S0133  10  12  12 
D04S0134  6  7  6 
D04S0135  6  9  7 
D04S0136  6  6  6 
D04S0137  7  7  7 
D04S0138  5  6  6 
D04S0139  6  7  7 
D04S0140  2  3  3 
D04S0141  5  6  5 
D04S0142  6  8  8 
D04S0143  5  5  5 
D04S0144  6  9  9 
D04S0145  5  7  7 
D04S0146  7  9  8 
D04S0147  7  8  7 
D04S0148  6  8  7 
D04S0149  7  9  9 
D04S0150  7  10  9 
D04S0151  11  12  11 
D04S0152  9  11  11 
D04S0153  5  7  6 
D04S0154  6  8  7 
D04S0155  4  4  4 
D04S0156  7  8  8 
D04S0157  4  5  4 
D04S0158  3  3  3 
D04S0159  1  1  1 
D04S0160  2  2  2 
D04S0161  3  4  4 
D04S0162  6  6  6 
D04S0163  4  5  5 
D04S0164  3  5  3 
D04S0165  6  10  8 
D04S0166  6  10  9 
D04S0167  6  6  6 
D04S0168  3  4  4 
D04S0169  6  6  6 
D04S0170  3  4  3 
D04S0171  4  7  6 
Conclusion
Weighting HE regression models by informativity is an appealing approach. However, some care is needed when applying this approach to real data. If phenotypes are weighted appropriately within families, this may result in substantial gain in statistical power [2, 3]. Instead of weighting phenotypes, an increase in power might also be obtained by weighting according to marker informativity. This approach has been successfully utilized in the context of metaanalyses in which studies have been weighted according to their informativity [6, 7]. The method of Jacobs et al. [4] combines both approaches. Jacobs et al. weighted individual sibpair families in the HE regression according to their marker informativity as indicated above. However, the asymptotic normality of the proposed test statistic was not shown.
In this paper, using simulated data available for GAW14, we demonstrated that the HE method with familywise weights according to marker informativity suffers from inflated type I error levels when we measure informativity as the distance between current IBD marker information and the IBD information for uninformative markers. If pvalues are computed asymptotically from weighted regression models only, the ttest statistic maybe distorted.
We therefore recommend the use of weighting functions in conjunction only with empirically computed pvalues until a theoretical solution to the detected problem is found. Furthermore, it needs to be investigated whether empirical p values in fact yield an increase in statistical power for weighted HE compared to the classical HE regression.
Abbreviations
 CI:

Confidence interval
 GAW:

Genetic Analysis Workshop
 HE:

Haseman and Elston
 IBD:

Identical by descent
Declarations
Acknowledgements
This work was supported by Deutsche Forschungsgemeinschaft grant ZI 591/121.
Authors’ Affiliations
References
 Haseman JK, Elston RC: The investigation of linkage between a quantitative trait and a marker locus. Behav Genet. 1972, 2: 319. 10.1007/BF01066731.View ArticlePubMedGoogle Scholar
 Amos CJ, Elston RC, Wilson AF, BaileyWilson JE: A more powerful robust sibpair test of linkage for quantitative traits. Genetic Epidemiol. 1989, 6: 435449. 10.1002/gepi.1370060306.View ArticleGoogle Scholar
 Sham PC, Purcell S: Equivalence between HasemanElston and variancecomponents linkage analyses for sib pairs. Am J Hum Genet. 2001, 68: 15271532. 10.1086/320593.PubMed CentralView ArticlePubMedGoogle Scholar
 Jacobs KB, GrayMcGuire C, Cartier KC, Elston RC: Genomewide linkage scan for genes affecting longitudinal trends in systolic blood pressure. BMC Genet. 2003, 4 (Suppl 1): S8210.1186/147121564S1S82.PubMed CentralView ArticlePubMedGoogle Scholar
 S.A.G.E.: Statistical Analysis for Genetic Epidemiology. 2004, 4.6: [http://darwin.cwru.edu/sage/]
 Loesgen S, Dempfle A, Gölla A, Bickeböller H: Weighting schemes in pooled linkage analysis. Genet Epidemiol. 2001, 21 (Suppl 1): S142S147.PubMedGoogle Scholar
 Dempfle A, Loesgen S: Metaanalysis of linkage studies for complex diseases: an overview of methods and a simulation study. Ann Hum Genet. 2004, 68 (Pt 1): 6983. 10.1046/j.15298817.2003.00061.x.View ArticlePubMedGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.