METHODS FOR ANALYZING PEDIGREE DATA WITH MANY PARAMETERS

该项目的广泛、长期目标是开发新的 用复杂模型分析家系数据的方法。与 目前基因数据的爆炸性增长，科学家们变得更加雄心勃勃 他们在努力定位疾病易感基因。而不是 满足于映射简单显性或隐性d，更 注意力集中在复杂的特征上，如糖尿病或 阿尔茨海默病。同时，多个标记上的数据可能 在单个分析中使用，要么是为了增加检测能力 或在找到连锁后对基因进行定位。这些 变化在数据分析中产生了两个独立但又相互关联的问题。 第一个问题是计算。因为适合的模型很复杂 对于疾病表型和/或多个标记的使用，只需 计算LOD分数可能非常耗时，有时 使用现有技术是不可行的。第二个问题是 统计推断。当前声明链接的标准是 LOD分数达到3.0或以上。虽然这一标准相当于 对于一个简单的特征，显然需要进行调整 当所拟合的遗传模型具有多个参数时，以及当多个 考虑了诊断方案。光是拿到Lod分数就应该 不能被认为是分析的结束。其他推理工具， 例如P值和后验概率，是需要的。拟议的研究 是针对上述两个问题提出的。从计算上讲， 高效的算法是通过结合几个 技巧。其中包括传统的剥离算法，重要性 抽样和吉布斯抽样。算法的开发不仅仅是为了 计算LOD分数，还用于获得P值和后验 赔率。此外，关键重组事件的概率还可以 作为奖金获得，这有时可以极大地帮助完成 理解数据。新的方法将使用这两种方法进行测试 模拟和真实的数据集。后者包括关于恶性疾病的数据集 黑色素瘤、阿尔茨海默病、一种罕见的糖尿病和骨质疏松症。

The broad, long-term objective of this project is to develop new methodology for analyzing pedigree data with complex models. With the current explosion of genetic data, scientists have become more ambitious in their effort to locate disease susceptibility genes. Instead of being satisfied with mapping simple dominant or recessive d , more attention is directed towards complex traits such as diabetes or Alzheimer disease. At the same time, data on multiple markers may be used in a single analysis, either to increase the power in detecting linkage or to localize the gene after linkage has been found. These changes created two separate, but related, problems in data analysis. The first problem is computations. Because of the complex models fitted to the disease phenotype and/or the use of multiple markers, simply computing the lod score can be extremely time consuming and sometimes infeasible using existing techniques. The second problem concerns statistical inference. The current criterion for declaring linkage is to have a lod score of 3.0 or above. While this criterion is quite appropriate for a simple trait, it is clear that adjustments axe needed when the genetic models fitted have many parameters and when multiple diagnostics schemes are considered. Getting the lod score alone should not be considered as the end of an analysis. Other inference tools, such as P-values and posterior odds, are needed. The proposed research is directed towards the two problems mentioned. Computationally, efficient algorithms are developed by combining the strengths of several techniques. These include the traditional peeling algorithm, importance sampling, and Gibbs sampling. Algorithms are not just developed for computing lod scores, but also for obtaining P-values and posterior odds. In addition, probabilities of key recombination events can also be obtained as a bonus, which can sometimes greatly assist the task of understanding the data. The new methodology will be tested using both simulated and real data sets. The later include data sets on malignant melanoma, Alzheimer disease, a rare form of diabetes, and osteoporosis.