An empirical comparison of case-control and trio based study designs in high throughput association mapping.

Motivated by high throughput genotyping technology, our aim in this study was to experimentally compare the power and accuracy of case-control and family trio based approaches for haplotype based, large scale, association gene mapping. We compared trio based and case-control study designs in different disease models, and partitioned the performance differences into separate components: those from the sample ascertainment, the effective sample size, and the haplotyping approaches. For systematic and controlled tests, we simulated a rapidly expanding and relatively young isolated population. The experiments were also replicated with real asthma data. We used computationally efficient methods that scale up to large amounts of both markers and individuals. Mapping is based on a haplotype association test for haplotypes of 1-10 markers. For population based haplotype reconstruction, we use HaploRec, and compare it to both a simple trio based inference and true haplotypes. Firstly and surprisingly, statistically inferred population based haplotypes can be equally powerful as true haplotypes. Secondly, as expected, the effective sample size has a clear effect on both gene detection power and mapping accuracy. Thirdly, the sample ascertainment method does not have much effect on mapping accuracy. Finally, an interesting side result is that the simple haplotype association test clearly outperformed exhaustive allelic transmission disequilibrium tests. The results suggest that the case-control design is a powerful alternative to the more laborious family based ascertainment approach, especially for large datasets, and wherever population stratification can be controlled.

Association analysis for quantitative traits by data mining: QHPM.

Previously, we have presented a data mining-based algorithmic approach to genetic association analysis, Haplotype Pattern Mining. We have now extended the approach with the possibility of analysing quantitative traits and utilising covariates. This is accomplished by using a linear model for measuring association. We present results with the extended version, QHPM, with simulated quantitative trait data. One data set was simulated with the population simulator package Populus, and another was obtained from GAW12. In the former, there were 2-3 underlying susceptibility genes for a trait, each with several ancestral disease mutations, and 1 or 2 environmental components. We show that QHPM is capable of finding the susceptibility loci, even when there is strong allelic heterogeneity and environmental effects in the disease models. The power of finding quantitative trait loci is dependent on the ascertainment scheme of the data: collecting the study subjects from both ends of the quantitative trait distribution is more effective than using unselected individuals or individuals ascertained based on disease status, but QHPM has good power to localize the genes even with unselected individuals. Comparison with quantitative trait TDT (QTDT) showed that QHPM has better localization accuracy when the gene effect is weak.

A second-generation association study of the 5q31 cytokine gene cluster and the interleukin-4 receptor in asthma.

We have analyzed a dense set of single-nucleotide polymorphisms (SNPs) and microsatellites spanning the T-helper cytokine gene cluster (interleukins 3, 4, 5, 9, and 13, interferon regulatory factor-1, colony-stimulating factor-2, and T-cell transcription factor-7) on 5q31 and the gene encoding the interleukin-4 receptor (IL4R) on 16p12 among Finnish families with asthma. As shown by haplotype pattern mining analysis, the number of disease-associated haplotype patterns differed from that expected for the 129Q allele polymorphism in IL13 for high serum total immunoglobulin (Ig) E levels, but not for asthma. The same SNP also yielded the best haplotype associations. For IL4R, asthma-associated haplotype patterns, most spanning the S411L polymorphism, showed suggestive association. However, these haplotypes consisted of the major alleles for the intracellular part of the receptor and were very common among both patients and controls. The minor alleles 503P and 576R have been reported to be associated with decreased serum IgE levels and changes in the biological activity of the protein, especially when inherited together. In the Finnish population, these two polymorphisms segregated in strong linkage disequilibrium. Our data support previous findings regarding L4R, indicating that 503P and 576R may act as minor protecting alleles for IgE-mediated disorders.

Mining associations between genetic markers, phenotypes, and covariates.

We used Haplotype Pattern Mining, HPM [Toivonen et al., Am J Hum Genet 67:133-45, 2000], for gene localization in Genetic Analysis Workshop (GAW) 12 isolate data. In HPM, association is analyzed by searching all trait-associated haplotype patterns. Data mining algorithms are utilized to make the search efficient. The strength of the haplotype-trait associations is measured by a linear model, into which a pre-seelected set of covariates is incorporated. Marker-wise patterns of association are used for predicting the disease gene location. Genome-wide scans of susceptibility genes for affection status as well as for the quantitative traits (Q1-Q5) were performed. First analyses were made with small sample sizes, 63-94 trios per trait, which is compared with a pilot study of a larger complex disease-mapping project. Subsequently, the analysis was repeated with approximately 600 cases and 600 controls per trait to give higher power to the analyses. With small sample sizes, only the susceptibility genes having the strongest effects on the traits could be localized. The larger sample size gave very good results: all susceptibility genes, except one, could be correctly localized. First experiments on candidate genes suggested that HPM is applicable even to fine mapping of mutations in DNA sequence.

Data mining applied to linkage disequilibrium mapping.

We introduce a new method for linkage disequilibrium mapping: haplotype pattern mining (HPM). The method, inspired by data mining methods, is based on discovery of recurrent patterns. We define a class of useful haplotype patterns in genetic case-control data and use the algorithm for finding disease-associated haplotypes. The haplotypes are ordered by their strength of association with the phenotype, and all haplotypes exceeding a given threshold level are used for prediction of disease susceptibility-gene location. The method is model-free, in the sense that it does not require (and is unable to utilize) any assumptions about the inheritance model of the disease. The statistical model is nonparametric. The haplotypes are allowed to contain gaps, which improves the method's robustness to mutations and to missing and erroneous data. Experimental studies with simulated microsatellite and SNP data show that the method has good localization power in data sets with large degrees of phenocopies and with lots of missing and erroneous data. The power of HPM is roughly identical for marker maps at a density of 3 single-nucleotide polymorphisms/cM or 1 microsatellite/cM. The capacity to handle high proportions of phenocopies makes the method promising for complex disease mapping. An example of correct disease susceptibility-gene localization with HPM is given with real marker data from families from the United Kingdom affected by type 1 diabetes. The method is extendable to include environmental covariates or phenotype measurements or to find several genes simultaneously.