New Bayesian algorithms for genome-wide association mapping

描述(由申请人提供)：全基因组关联研究很有希望揭示人类复杂疾病背后的遗传结构。疾病变异体通常是非孟德尔变异体，表现出低外显性，对疾病个体影响很小，但彼此之间和环境之间以未知的方式相互作用。随着最近的高通量测序技术的发展，在基因组水平上产生了更多的数据，不仅包括遗传变异，还包括个体水平的调控元件。已知调节因子在序列变异和表型多样性之间相互作用并起中介作用。因此，多变异疾病图谱对于未来的全基因组关联研究变得更加有趣和重要。人们还希望，通过收集人类基因组中的所有变异，我们可以识别真正的致病变异，这样就可以在识别的位置精确地进行功能评估和验证实验，以真正揭示它们对疾病的生物学机制。识别多变量关联是一项极具挑战性的工作。目前的算法仍然非常有限。特别是，高通量测序数据现在在疾病研究中经常产生。这些完整的变种是高度依赖的，现有的方法对此有很大的计算困难，因此很难准确定位真正的疾病变种。从罕见变异中发现疾病关联也是非常具有挑战性的，然而，罕见变异在人类基因组中更为丰富，可能是人类复杂疾病的主要贡献者。我们建议开发先进的算法来解决上述问题。我们将开发先进的算法来提高全基因组多变量作图的能力和计算效率。我们还提出了在一致的全概率模型下联合测试常见和罕见变量的通用方法。我们的方法自动对变种进行联合测试，考虑依赖性，纳入生物学先验，并识别致病变种。我们通过非参数贝叶斯技术进一步扩展了这些方法，以便在疾病地图中整合公共数据库的各种来源。我的新算法将极大地增强研究人员分析高通量遗传和基因组数据的能力。该软件将通过PI的网站和设在宾夕法尼亚州立大学的Galaxy系统免费分发给社区。

DESCRIPTION (provided by applicant): Genome-wide association studies hold great promises to reveal the genetic architectures underlying human complex diseases. The disease variants are often non-Mendelian, demonstrating low penetrance and little effects to the disease individually, but interacting with each other and environments in unknown ways. With recent high-throughput sequencing technology, much more data are generated in the genome-scale, including not only genetic variants, but also regulatory elements at the individual-level. Regulatory factors are known to interact and act as mediators between sequence variation and phenotypic diversity. Multi-variant disease mapping therefore becomes more interesting and important for future genome-wide association studies. It is also hoped that, by collecting all variants in the human genome, we could identify the true causative variants, such that functional evaluation and validation experiments can be precisely developed at the identified sites to truly reveal their biological mechanisms to the disease. Identifying multi-variant association is extremely challenging. Current algorithms are still very limited. Particularly, high throughput sequencing data are now routinely generated in disease studies. These complete variants are highly dependent, for which existing methods have substantial computational difficulties and thus make it extremely difficult to pinpoint the true disease variants. It is also very challengingto detect disease associations from rare variants, which are however more abundant in the human genome, and could be the main contributor to human complex diseases. We propose to develop advanced algorithms to tackle the above problems. We will develop advanced algorithms to improve the power and the computational efficiency for whole genome multi-variant mapping. We also propose generalized methods to jointly test common and rare variants under a coherent full probabilistic model. Our approach automatically group variants for joint testing, account for dependence, incorporate biological priors, and identify causative variants. We further extend the methods via non-parametric Bayesian techniques to integrate various sources of public databases in disease mapping. My new algorithms will greatly enhance researchers' capability to analyze high-throughput genetic and genomic data. The software will be freely distributed to the community through the PI's website and the Galaxy system hosted at Penn State.