Context-Specific Gene Regulations in Cancer Gene Expression Data
The idea of context-specificity in the biological system is an important one as research goes forward into the new decade. To those who approximate or model biological systems, especially with formal mathematical representations, the fact that the true underlying biological system is not homogeneous but is actually the combination of many overlapping and interconnecting “contexts” must be considered.
Our work continues to date, but for now this page contains source code and data files supplementing our submission accepted to the 2009 Pacific Symposium on Biocomputing entitled, “Context-Specific Gene Regulations in Cancer Gene Expression Data,” by Ina Sen, Michael P. Verdicchio, Sungwon Jung, Robert Trevino, Michael Bittner, and Seungchan Kim.
Introduction
The introductory section of our paper gives a succinct overview of contextual genomic regulation and it is reproduced here:
Under normal conditions, a cell maintains a specific state by tightly controlling various molecules using a variety of regulatory mechanisms. In the face of environmental changes, a cell adjusts its regulatory mechanisms accordingly. Mutation or other types of damage that alter these regulatory mechanisms may erode this control and cause the cell to transition into another state significantly different from the prior normal state. If the normal state is taken to be “healthy” and the altered state is taken to be “tumor”, for example, the regulatory functions must have been altered in significant ways to arrive at the “tumor” state. Since the way the system interprets and acts upon certain inputs is altered, we say that there is a change in cellular context.
Although a tumor state of the cell is different from normal, the continuing proliferation and survival of cancer shows that such a state is indeed steady and maintained by complex regulatory behavior. If one can learn from the contextual information which regulating mechanisms differ from context to context, then one can potentially discover the mechanisms that initiate and maintain complex, hard-to-treat diseases, such as cancer.
High throughput data collection methods, including gene expression microarrays, provide vast amount of data to study various aspects of cellular processes. Many methods and techniques exist to discern and model the regulatory behavior of cells, and each certainly has distinct advantages and disadvantages. For instance, traditional clustering approaches like k-means or hierarchical clustering can help group samples or genes, revealing possible novel subtypes of diseases or subclasses of molecular functions. Bayesian networks have been employed as models of genomic regulation; however they inherently assume homogeneity of samples and thus cannot model different cellular contexts, a serious limitation in non-homogeneous disease like cancer.
In this paper, we will first describe a mathematical model of contextual genomic regulation, a method based upon that model to identify cellular contexts, and then propose a novel method to construct context-specific gene regulatory networks. We apply the context mining method to gene expression data collected from a broad spectrum of cancer patients to reveal the modular and context-specific structure of gene regulatory networks hidden within the data. Finally, we conclude with the future direction of our work.
Overview of Paper
Manuscript (pdf)
Flowchart of the paper (png)
Graphic overview of context mining with notation from paper (png)
Previous paper on Cellular Context Mining (link)
Previous paper on Context-Specific Gene Regulatory Networks (link)
Paper on Contextual Genomic Regulation Model (link)
Software, Scripts and Data Files
ExPattern (link)
ExPattern project file (exp, right-click and save) (put in same directory as data set and xml)
TargetNow dataset for ExPattern (zip) (extract and put in same directory as project file and xml)
Context Mining Results for ExPattern (zip) (extract and put in same directory as project file and data)
Matlab scripts for determining sample association scores (zip)
Input files for sample association Matlab scripts (zip)
Graphs and Figures
