home.uchicago.edu/~abney/abney_web/Publications_files/han2011a.pdf In this work we developed a method, IBDLD, that can rapidly estimate IBD sharing between pairs of individuals related through an arbitrary pedigree given dense genotype data. The problem of estimating IBD in large pedigrees given multipoint genotype data has been a particularly vexing one to geneticists resulting in a variety of pedigree splitting strategies. All such approaches, however, necessarily entail a loss of information which can either lead to a significant reduction in power [Dyer et al., 2001] or increase in false positives [Newman et al., 2001] when performing mapping. These difficulties have been compounded recently with the wide use of genotyping chips with high-density genotyping data. The presence of LD in such data has the consequence of rendering the standard HMM inappropriate. IBDLD overcomes both difficulties and is computationally efficient enough to use genomewide on samples of at least several hundred related individuals. When using an HMM model that does not incorporate LD, the two methods NoLD and NoLD-S represent two extreme SNP pruning strategies, no pruning and severe pruning, respectively. Other, more sophisticated pruning strategies could be used which would result in accuracy intermediate between the two extremes, while keeping computation time to a minimum. Our results indicate, however, that regardless of the pruning strategy used, accuracy will be far worse thanmethods that incorporate LD into the model, at the cost of additional computation time. We explore two different models for including the background LD. Results from the two models indicate that they generally perform similarly, though the LD-RR method appears to be somewhat more robust than LD-20 in the presence of missing genotypes and genotyping error. Additionally, the LD-20 method requires phased haplotypes in the panel of individuals from whom the background LD will be modeled. Although if such a panel is available this does not pose any difficulty, it may often be the case that there is no such suitable data. Using a panel that accurately represents the LD in the study sample is critical to the accuracy of the method. In the analysis of the Hutterite data one might expect that the HapMap CEU population would provide an accurate representation of the LD as the Hutterites are a European Caucasian-derived population. Using this panel, however, resulted in significant bias in the estimates of IBD sharing. Instead, using the same sample in which we wish to estimate IBD to also model the LD gave highly accurate estimates. In the context of estimating IBD in a sample of individuals related through an arbitrary pedigree, then, it becomes necessary to phase all the individuals. Standard methods for phasing individuals with dense genomewide data typically assume the subjects are unrelated. Phasing within a pedigree, with the expectation that Mendelian inheritance laws will be obeyed, is a laborious and potentially error prone task. Approaches such as the one used by Kong et al. [2008] may be promising in this respect. The difficulty with phasing in this context results in our LD-RR method, which requires only unphased genotypes in the LD modeling panel, being advantageous. In spite of IBDLD showing sensitivity to correctly modeling the background LD, it is highly robust to pedigree misspecification. In our analysis of the Hutterite data, when a sample was duplicated into an incorrect person in the pedigree, the estimates of IBD were essentially unchanged, even though the estimates may have been distant from their expected value based on the purported positions of the individuals in pedigree. Although a tool such as PREST remains useful at detecting individuals who might have a misspecified relationship with other pedigree members, it has difficulty in identifying the actual relationship much of the time. IBDLD, on the other hand, can still accurately estimate the actual amount of IBD sharing and, hence, suggest a very likely true position in the pedigree. Nevertheless, PREST is extremely fast and maintains significant utility as a screening tool. The method presented here would be particularly useful as a follow-up for pairs that appear to be sharing anomalously relative to their known pedigree locations. The robustness to misspecified relationships is the result of the high level of information in very dense SNP data. The pedigree connecting a pair of individuals is used to both estimate an expected level of IBD sharing and a transition rate between IBD states. When the genotype information is highly informative toward IBD, the information from the pedigree contributes a relatively small part. This leads to accurate estimates even when the pedigree is incorrect. A particularly useful implication of this is that it may be possible to obtain highly accurate estimates of IBD even when the pedigree is unknown. Though it will always be more accurate to use pedigree information, in populations where the individuals are likely to be highly related but where the genealogy is unavailable, it is probable that a modification of the method will still be able to reasonably estimate IBD sharing. We are currently exploring this idea. A particular use of IBDLD is in the mapping of complex traits. Family studies that can combine both linkage and association information may prove effective at helping to uncover some of the ‘‘missing heritability’’ that plagues the mapping of common traits. The ability to use large samples of related people with dense SNP data to obtain IBD estimates that can then be used in a mixed model approach [Almasy and Blangero, 1998; Kang et al., 2010] may increase power to detect QTL. Additionally, using actual rather than expected IBD sharing may also lead to greater insight into genetic architecture by not only giving better estimates of overall heritability of traits but also allowing one to assign heritability to either particular chromosomes or chromosomal regions, e.g. Visscher et al. [2007].
Posted on: Wed, 16 Oct 2013 17:24:05 +0000