• 统计研究中心
当前位置: 首页 > 系列讲座 > 正文

香港科技大学杨灿副教授: MR-APSS: a unified approach to Mendelian Randomization accounting for pleiotropy, sample overlap and selection bias using the genome-wide summary data

光华讲坛——社会名流与企业家论坛第 5758 期

(线上讲座)

主题MR-APSS: a unified approach to Mendelian Randomization accounting for pleiotropy, sample overlap and selection bias using the genome-wide summary data

主讲人香港科技大学杨灿副教授

主持人统计学院 林华珍教授

时间2020年7月17日(周五)下午4:00-5:00

直播平台及会议ID腾讯会议,588 795 997

主办单位:统计研究中心、数据科学与商业智能联合实验室和统计学院 科研处

主讲人简介:

杨灿博士现为香港科技大学数学系副教授,健康数据分析中心主任,大数据研究院教授成员,与金融科技硕士项目机器学习方向负责人,WeGene公司科学顾问。他分别于2003年和2006年在浙江大学获得工学学士学位和工学硕士学位,并于2011年在香港科技大学获得电子计算机工程博士学位。他是耶鲁大学的博士后(2011-2012)和副研究员(2012-2014)。他的研究领域专注于统计方法的开发以及计算工具在大规模数据分析中的应用(例如BOOSTGPA)。他的研究论文发表在一系列高影响力的期刊上,Annals of Statistics, Bioinformatics, IEEE Transactions on Pattern Analysis and Machine Intelligence, Nature Communications, PLoS Genetics, Proceedings of the National Academy of Sciences, 以及 The American Journal of Human Genetics。基于杨灿博士对数据分析方法和工具的贡献,他荣获2012年香港青年科学家一等奖。截至20206月,杨博士的工作已被引用3040次,h指数为26i10指数为37。杨灿博士获得香港政府创新技术基金的支持,与WeGene公司建立起产业合作。详情请见其个人主页:http://www.math.ust.hk/people/faculty/profile/macyang/

内容提要:

Inferring the causal relationship between a factor (exposure) and a phenotype of interest (outcome) is essential in biomedical research. Examples of causal questions include: Is high-density lipoprotein cholesterol (HDL-C) a protective factor for coronary artery disease (CAD)? Is prenatal exposure to smoking a risk factor of disruptive behavioral disorders in children? A significant challenge for answering these causal questions is the confounding issue: association between exposure and outcome can be induced by confounding factors, even though there is no causal relationship. To address challenges in causal inference, randomized controlled trials (RCTs) are often regarded as the gold standard. However, RCTs can be very expensive and sometimes even infeasible and unethical (e.g., random allocation to prenatal smoking). To overcome the limitations of RCTs, Mendelian randomization (MR) was introduced to mimic RCTs to perform causal inference. According to Mendelian Laws of Inheritance, genotypes are randomly assigned from one generation to next generation and they are unaffected by confounding factors. Therefore, genotypes can be used as instrumental variables (IVs) to eliminate the influence of confounding factors. However, MR has several unique challenges due to the complexity of human genetics, including polygenicity, pleiotropy, potential bias induced by data pre-processing and summary-statistics sharing. In this talk, we introduce a statistical approach, MR-APPS, addressing the unique challenges in MR. We have applied MR-APSS to GWAS summary statistics of 29 traits and 24 traits from UK BioBank and genomics consortia, respectively. The analysis results indicate that MR-APSS can provide more plausible causal findings with high reliability. This is a joint work with Xianghong Hu, Jia Zhao, Yang Wang, Heng Peng, and Xiang Wan.

在生物医学研究中,推断因子(暴露)与感兴趣表型(结果)之间的因果关系至关重要。因果关系问题的例子包括:高密度脂蛋白胆固醇(HDL-C)是冠状动脉疾病(CAD)的保护因子吗?产前吸烟是儿童破坏性行为障碍的危险因素吗?回答这些因果问题的一个重大挑战是令人困惑的问题:即使没有因果关系,混杂因素也可以诱发暴露和结果之间的关联。为了解决因果推理方面的挑战,随机对照试验(RCT)通常被视为黄金标准。但是,随机对照试验非常昂贵,有时甚至是不可行和不道德的(例如,随机分配给产前吸烟)。为了克服RCT的局限性,引入孟德尔随机(MR)来模拟RCT进行因果推理。根据孟德尔遗传定律,基因型是从一代到下一代随机分配的,并且不受混杂因素的影响。因此,基因型可以用作工具变量(IV),以消除混杂因素的影响。但是,由于人类遗传学的复杂性,MR面临一些独特的挑战,包括多基因性,多效性,由数据预处理和汇总统计信息共享引起的潜在偏差。在本演讲中,我们介绍一种统计方法MR-APPS,以解决MR中的独特挑战。我们已经将MR-APSS应用于来自英国生物银行和基因组学联盟的29个性状和24个性状的GWAS摘要统计中。分析结果表明,MR-APSS可以提供更高可信度的更合理的因果发现。本研究是与Xianghong HuJia Zhao, Yang Wang, Heng Peng, and Xiang Wan共同完成的。


上一条:英国伯明翰大学雷云文博士:基于随机梯度下降的统计学习

下一条:香港大学 李国栋教授:High-dimensional vector autoregressive time series modeling via tensor decomposition