※ [本文转录自 NCTU-STAT96G 看板] 作者: cyz ( ) 看板: NCTU-STAT96G 标题: 1130演讲＠交大统研所 时间: Wed Nov 21 13:25:39 2007 国立交通大学、清华大学 统计学研究所 专题演讲 题　目：Evolution of Multiple Hypothesis Testing in Genetic Studies 主讲人：萧朱杏教授 (台湾大学公共卫生学系) 时　间：96年11月30日(星期五)上午10:40-11:30 (上午10:20-10:40茶会於交大统计所428室举行) 地　点：交大综合一馆427室 Abstract The recent advancement of biotechnology and the announcement of human genome have made the genetic association studies possible and easier to conduct. It also opens a new era in statistical sciences. The challenge of testing simultaneously a large number of hypotheses in genomics has been receiving considerable attention. It differs from the traditional multiple comparisons performed after a significant overall test. The small sample size, the large number of tests, and the correlation within data have made the issue even more complex. Statisticians have tackled the problem from different angles. Several algorithms focused on selection of ordered p-values via a bottom-up or top-down procedure, possibly under a controlled false discovery rate (FDR); while some recommended a multi-stage procedure using data from different subjects at every stage, or starting with a smaller group of individuals and then augmenting the data in each of the rest stages. Recently, several researchers proposed to replace the standard null normal distribution with other dispersed ones for the statistics under the null or the alternative hypothesis. Alternatively, the use of Bayesian mixture model considers simultaneously the character of classifying the hypotheses and the general dependence among data. In this talk, we will discuss briefly the current development, introduce the Bayesian approach, and highlight some possible directions. (This is a joint work with Yu-Chung Wei and Shu-Hui Wen) 敬 请 公 布 欢 迎 参 加 --

※ 发信站: 批踢踢实业坊(ptt.cc) ◆ From: 140.113.114.28 --

※ 发信站: 批踢踢实业坊(ptt.cc) ◆ From: 140.113.114.185