The 40th In Silico Megabank Research Seminar will be held on Friday, December 6.
This Time, we will be welcoming Dr. Osamu Komori, The Institute of Statistical Mathematics as our lecturer, and he will be speaking on “Discussion of Asymptotic Properties of Generalized t-statistics and Its Application to Actual Data Analysis .”
・Date/Time: December 6(Fri.) 17:00‐18:30
・Venue: Conference Room 1(2nd Floor), Tohoku Medical Megabank Organization
・Title:Discussion of Asymptotic Properties of Generalized t-statistics and Its Application to Actual Data Analysis
・Lecturer: Osamu Komori ( The Institute of Statistical Mathematics)
＊This lecture is transferable as a class in the medical research-related lecture course.
・Abstract: In recent years, search for variable quantities (makers) that are valid for high-dimensional data analysis used at settings such as clinical medicine or various diagnosis of illness has become more and more important. For analysis on high-dimensional continues variable such as data of gene expression levels, often t statistics and c statistics (AUC) are used at the phase narrowing down the variables. In this research, we focus on the t statistics, one of test statistics, and consider the application of t statistics to classification problems while taking account of multivariate linear combination instead of univariate. By considering U Function, a generating function to t statistics, we can discuss generalization of various t statistics. This research revealed close relation among t statistics, c statistics (AUC), Fisher Linear Discriminant, and Kullback-Libler divergence. In addition, the result suggests Lasso type method which L1 penalty is added to generalized t statistics as an example of application of this method to actual data analysis. One feature of this method is that selection of other variables is possible after fixing variables that are recognized as valid beforehand (such as variables that strong relation to risk of development). Its usability is to be examined with simulation and actual data analysis. In addition, we would like to consider applicable statistical method on discrete data such as SNP in the future.
・Organizer: Gen Tamiya, Masao Nagasaki
Access : http://www.megabank.tohoku.ac.jp/english/info/access.html