Document Actions

(04-07-05)Ker-Chau Li

Statistics Department, UCLA

Title: An integrated system for co-exploring aggregated gene expression database, protein complex, genetic markers, drug sensitivity profiling and genomic knowledgebase

Abstract: Gene expression data can be thought of as a matrix with N rows and M columns. Each row represents the expression profile of a gene under M conditions. N is the number of genes, which is about 6000 for yeast, and can be over 30,000 for human. For the microarray experiments that we are interested in, M is typically greater than 50. One central question in bioinformatics is how to help biologists deduce interesting information from such a matrix. Motivated by the rationale that genes with similar profiles are likely to be functionally associated, methods such as hierarchical clustering, principal component analysis, self-organization map and other variants have been widely used. They offer biologists valuable genome-wide portraits of how clusters of genes may be coregulated. But such approaches have a limitation. As it turns out, the majority of genes do not fall into the detected clusters. If one has a gene of primary interest in mind and cannot find any nearby clusters, what additional analysis can be conducted? In this talk, I will show how to address this issue via the newly developed statistical notion of liquid association(1,2,3). I will describe several on-going projects in my lab (http://kiefer.stat.ucla.edu/lap) , using this and other statistical ideas to distil information from a web of aggregated genomic knowledgebase and data sources at multi levels, including gene ontology, protein complexes, genetic markers, drug sensitivity, and disease candidate genes. The examples include arginine biosynthesis/urea cycle regulation in yeast, expressional interaction between Alzheimer disease genes such as APP, PSEN, APOE, and BACE, correlating the drug sensitivity profiles for anticancer compounds such as methotrexate (MTX) and taxol with their molecular target gene expression, an exploration of the poorly characterized human prion (PRNP). Finally I will also discuss an application to multiple sclerosis study, a newly launched project with researchers at Human Genetics Department UCLA.

Comment:

function association $\Rightarrow$ expr correlation. Not true.
Transcription factors can be either activator or repressor
because X,Y are normalized to have mean 0.
$E(XY)=E(E(XY\vert Z))=Eg(Z)$
$LA(X,Y\vert Z) = Eg^{'}(z)=Eg(Z)Z=E(XYZ)$ High and low LA scores are interesting. They represent the strong relationship between XY and Z.
?normal score transformation
people are all starting to look at relationship beyond 2-element. But first step is 3-element. Another example is also from ucla[2].
extend X,Y to 2+ element. $LA(f(X),g(Y)\vert Z)$ , f(X) and g(Y) are two groups of genes.
extend Z to 2+ element.
look at the data of the genes whose relationship you've known. From biology to data and then generalize data to biology.
Don't throw away negative correlations. (one subsection below5.5)