STATISTICAL METHODS FOR GENE MAPPING

受到人类基因组巨大的数据处理需求的刺激 项目，遗传学正在迅速成为一门计算科学。这 该提案解决了人类基因中出现的一些计算问题 映射。家庭研究、精子分型和辐射杂交都在现场 构造高分辨率中的独特建模计算挑战 基因图谱。即使拥有超快的计算机，良好的算法设计和 软件开发势在必行。 隐马尔可夫链理论提供了一个通用的框架 在这三种基因作图策略中的可能性评估。什么时候 结合EM算法进行最大似然估计，FAST 评估程序是可能的。同样的优势很可能会在 进化树的重建。因为估计必须是 对于要映射的特定顺序的基因座，还存在 增加了调查大量候选订单的复杂性。 识别最佳订单的更好的贝叶斯和启发式技术 需要被开发。最后，较好的遗传随机模型 重组显然在家族研究和精子研究中都很重要 打字。 等位基因集中和蒙特卡罗马尔可夫链是另一个相关的 计算设备。等位基因集中在一个局部的 系谱可能消除链接中的一些主要计算瓶颈 计算。我们正在进行的关于Metropolis算法的谱系工作 高级语言独立性测试的交换算法分析与研究 维列联表也展示了一些伟大的前景 模拟方法适用于遗传学。这些项目的进一步发展 模拟工具由各种应用程序保证，包括仅受影响的应用程序 连锁分析的方法，最可能的基因型的鉴定 家系中的载体，以及对Hardy-Weinberg和连锁的测试 多位点群体数据中的平衡。 这一建议为上述研究提出了几条雄心勃勃的途径。 区域。我们的意图是追求这些最有希望的线索，同时 机会主义地关注遗传学的新发展 其他问题。生产可用的软件将是一种自然的 我们调查的结果。

Spurred by the enormous data processing demands of the Human Genome Project, genetics is quickly becoming a computational science. This proposal addresses some of the computational issues arising in human gene mapping. Family studies, sperm typing, and radiation hybrids all present unique modeling computational challenges in constructing high-resolution gene maps. Even with ultra-fast computers, good algorithm design and software development are imperative. The theory of hidden Markov chains provides a common framework for likelihood evaluation in these three gene-mapping strategies. When combined with the EM algorithm for maximum likelihood evaluation, fast estimation procedures are possible. The same advantages may well accrue in the reconstruction of evolutionary trees. Because estimation must be carried out for a particular order of the loci to be mapped, there is also the added complexity of investigating a large number of candidate orders. Better Bayesian and heuristic techniques for identifying the best order need to be developed. Finally, better stochastic models of genetic recombination are obviously important in both family studies and sperm typing. Allele lumping and Monte Carlo Markov chains are other pertinent computational devices. Allele lumping done on a local basis within a pedigree may eliminate some the major computational bottlenecks in linkage calculations. Our ongoing work on the Metropolis algorithm in pedigree analysis and on exchange algorithms for testing independence in high- dimensional contingency tables also demonstrates some of the great promise simulation methods hold for genetics. Further development of these simulation tools is warranted by applications as diverse as affecteds-only methods of linkage analysis, identification of the most probable genotype vector in a pedigree, and testing for Hardy-Weinberg and linkage equilibrium in multilocus population data. This proposal suggests several ambitious avenues for research in the above areas. Our intention is to pursue the most promising of these leads while keeping an opportunistic eye on new developments in genetics for additional problems. Production of usable software will be a natural outgrowth of our investigations.