Li, Ker-Chau 李克昭

特聘研究員 / 本院統計科學研究所

Email: Email住址會使用灌水程式保護機制。你需要啟動Javascript才能觀看它


  • Ph.D., Statistics, 1981, University of California, Berkeley. (Advisor: Jack Kiefer)
  • B.S., Mathematics, 1975, National Taiwan University
  • 2006-present: Distinguished Research Fellow, Institute of Statistical Science, Academia Sinica
  • 2006-2012: Director, Institute of Statistical Science, Academia Sinica
  • 2000-2002: graduate vice chair, Statistics Department, UCLA
  • 1999-present: Professor, Statistics Department, UCLA
  • 1989-present: Professor, Mathematics Department, UCLA
  • 1984-1989: Associate Professor, Mathematics Department, UCLA
  • 1981-1984: Assistant Professor, Statistics Department, Purdue University


  • 2012 Academician, Academia Sinica
  • 2003 Medallion Lecturer, IMS; 1993 Guggenheim Fellow;
  • 1991 NSF/ASA/NIST fellow; 1990 JASA theory and methods Editor's invited speaker in Joint Statistical Meetings;
  • 1989 IMS Fellow; 1981 elected member of Phi Beta Kappa;
  • 1981 B. Friedman Memorial Prize in Applied Mathematics, U.C. Berkeley.


Bioinformatics, systems biology, lung cancer studies, High dimensional data analysis, Large ensembles of time series, Medical image analysis, Machine learning, Statistical graphics, Bayesian computation, Regression, Censoring, Experimental design, Survey sampling.


Li is best known for introducing sliced inverse regression (SIR) and principal Hessian direction (PHD), two fundamental dimension reduction methods for high dimensional data analysis. Starting from 2000, his research interest turned to the emerging field of computation/mathematics/statistics in genome biology. In 2002, he published a paper in Proceedings of Academy of Science, featuring the novel method of liquid association (LA) for microarray gene expression analysis. He is currently leading a research group in UCLA and in Academia Sinica to continue the development of methods for utilizing multiple sources of gene expression profiling, genetic markers, complex disease phenotypes and traits. A website offers on-line computation based on LA and related methods for gene expression studies. He is also collaborating with Dr. Pan Chyr Yang of National Taiwan University and his colleagues on integrative cancer biology.

Liquid association. (a) Association between genes X and Y as mediated by gene Z. When gene Z is expressed at the high level (red), a positive correlation between X and Y is observed. The association changes as the expression of Z is lowered. It eventually becomes a negative trend (green). There are two basic ways (shown in panels b and c) to apply the liquid association (LA) scoring system to guide a genome-wide search. (b) When two genes X and Y are given, compute LA score LA(X, Y|Z) for every gene Z first and then output a short list of high score genes Z1, Z2, and so on. (c) When only one gene X is given, compute LA score LA(X, Y|Z) for every pair of genes X,Y first and then output a short list of high score gene pairs Y1,Z1, Y2,Z2, and so on. Li et al. Genome Biology 2007 8:R205 doi:10.1186/gb-2007-8-10-r205


Liquid association. Correlation is a simple and powerful method for analyzing gene expression data.Two genes with positively correlated expression profiles are likely to be functionally associated. They may participate in the same biological process. However, functionally associated genes may not be correlated for a variety of reasons. For instance, they may not be regulated at the transcription level. Another common situation is that most genes have multiple functions. Depending on the cellular needs, co-expressed genes may become uncorrelated or even turn into contra-expressed. Liquid association (LA), as opposed to "steady" association, is designed to quantify the change of correlation between two genes as a relevant cellular state variable changes. There is no need to specify the state variable explicitly. A highlighted example is the elucidation of gene expression for the urea cycle in yeast. This pathway controls both the biosynthesis and degradation of the amino acid, arginine. LA is able to find the correct genes which have to be turned on, as well as the correct genes which have to be turned off at the same time, so that any wasteful immediate degradation of newly synthesized arginine can be avoided.


Sliced inverse regression (SIR). Many statistical methods are known to suffer from curse of dimensionality? they break down easily when dealing with high dimensional data. How to reduce dimensionality is a long-standing issue. Ad hoc methods such as principal component analysis and partial least squares have been advocated. Yet the associated issue about possible nformation loss due to improper dimension reduction is rarely addressed. The SIR methodology helps reshape this area by presenting an effective dimension reduction framework for theorizing both issues.


Principal Hessian direction. Another effective dimension reduction method.