Biclustering Methods for Microarray Data Analysis

Speaker: Kevin YIP

Abstract

With the advent of microarray technology, the activity of thousands of genes can be recorded simultaneously. Putting together the expression profiles of the genes under different conditions, the resulting data can be viewed as a large matrix, where each row corresponds to a gene and each column corresponds to a condition. There is a huge number of undergoing research efforts that try to dig out biological knowledge from these large matrices. While directly applying traditional machine learning or data mining methods on the data has been successful in some cases, it is more common to see learning results that have no apparent biological meaning. This may due to the special data characteristics of the microarray datasets, which do not fit well into the traditional data models. In the case of clustering, some parties have suggested that clusters may be better represented by submatrices that involve only subsets of rows and columns. The corresponding algorithms cluster both genes and conditions simultaneously. The problem, called biclustering, highly resembles the projected and subspace clustering problems studied by the database community in recent years. Interestingly, the approaches taken by the two communities are fundamentally different.

In this talk, I will introduce some biclustering methods proposed specifically for microarray data analysis. Some brief descriptions of the models and algorithms will be given, and the approaches will be compared based on some high-level criteria. I will also suggest some possible further research topics in biclustering analysis on microarray data.

Referred Papers