Bootstrap-based testing of rare sequence variants using family data

描述（由申请人提供）：我们现在有大量的测试库，用于使用不相关个体的基因型来检测疾病与基因组区域中罕见变异之间的关联。然而，只有其中最简单的方法被扩展到家庭数据。然而，使用相关病例的病例对照测试比仅基于不相关病例的测试更有效，特别是对于罕见的变异。权力增益反映了罕见的因果变异的受影响亲属的丰富。提高功率是至关重要的，因为大多数破坏性变异在人群中发生的频率非常低，而且需要大量的样本量和外部生物信息来检测与疾病的关联。病例和对照的多位点基因型之间基于生物学的对比可能是复杂的，在受试者基因型之间存在相关性的情况下，需要简单、灵活的方法来推断其零分布。我们提出了一种新的方法，将所有病例-对照关联试验扩展到所有受试者，而不管他们的家谱关系如何。新方法以一种新颖的方式使用Efron的自举，涉及对受试者相关的基因型数据进行“去相关”以允许自举重采样，然后对自举数据进行“重新相关”以推断检验统计量的零分布。目的1将使用模拟来验证新的准引导（QB）方法，该方法使用家族数据来识别疾病与复杂基因型组合的关联。这一目标包括：i)评估QB测试对家庭数据的1型误差和功效，与之相比：a)同样的测试适用于不相关的受试者；b)在可用的情况下，对家庭数据进行封闭式基于高斯的测试；ii)将QB方法扩展到包含群体结构和隐相关性的数据，这些数据必须估计成对受试者之间的相关矩阵；Iii)处理缺失的基因型数据。目标2将应用

DESCRIPTION (provided by applicant): We now have a large arsenal of tests for association between disease and rare variants in genomic regions using the genotypes of unrelated individuals. However only the simplest of them have been extended to family data. Yet case-control tests using related cases are more powerful than tests based only on unrelated cases, particularly for rare variants. The power gain reflects enrichment of affected relatives for rare causal variants. Increased power is critical because most damaging variants occur at very low frequencies in human populations, and huge sample sizes and external biological information will be needed to detect associations with disease. Biologically-based contrasts between the multi-locus genotypes of cases and controls are likely to be complex, and simple, flexible methods are needed to infer their null distributions in the presence of correlation among subjects¿ genotypes. We propose a new way to extend all case-control association tests to all subjects, regardless of their genealogical relationship. The new method, which uses the bootstrap of Efron in a novel way, involves ¿de-correlating¿ subjects¿ correlated genotype data to allow bootstrap resampling, and then ¿re-correlating¿ the bootstrapped data to infer the null distribution of the test statistic. Aim 1 will use simulations to validate the new Quasi-bootstrp (QB) method for using family data to identify associations of disease with complex combinations of genotypes. This aim includes: i) assessing the type-1 error and power of QB tests for family data in comparison to: a) the same tests applied to unrelated subjects; and b) closed-form Gaussian-based tests for family data when available; ii) extending the QB method to data containing population structure and cryptic relatedness, for which the correlation matrix between pairs of subjects must be estimated; iii) dealing with missing genotype data. Aim 2 will apply the QB method to cancer family data to evaluate its performance on functional genetic units containing known carcinogenic variants. This includes testing for BRCA1 and BRCA2 association with breast cancer in affected and unaffected subjects from families in the Breast Cancer Family Registry (BCFR) and testing for HOXB13 association with prostate cancer in the International Consortium on Prostate Cancer Genetics (ICPCG). Aim 3 will develop freely-available software to implement the QB method for existing multi-locus case-control association tests. This software will include methods for handling missing genotype data for some subjects at some markers. The software will allow users with data from related and unrelated subjects to evaluate associations with disease using any of the existing tests currently available only for unrelated subjects. If validated, the proposed QB method would provide a major addition to our tools for next-generation sequence data by analyzing those most likely to carry causal disease variants, while building on the known strengths of the bootstrap. These include ease of use, robustness, and versatility for a large variety of applications. With the computing resources now routinely available, the proposed method can be implemented quickly and easily. Narrative: Sequencing the genomes are many people is now cost-effective, and it may help us finds the groups of genes that cause chronic diseases such as cancer. However evidence now suggests that many very rare variants may act in concert to cause such disease, and unraveling the new clues will require evaluating the genomes of diseased individuals from families with multiple cases of the disease. We propose a simple way of applying any of the new tests to such families, which should increase their efficacy.