Multidimensional Signal Models for Microarray Data Biclustering

Microarrays or DNA chips can efficiently study thousands of genes under many conditions. Biclustering algorithms classify genes and conditions at the same time and can overcome limitations of conventional clustering methods. Existing biclustering algorithms often use heuristic approaches to avoid exponential computational complexity. Thus there is a clear need to develop more powerful biclustering methods. In this project, novel and robust biclustering techniques based on multidimensional signal models will be investigated. Effective techniques will be studied to detect and separate various types of biclusters, design efficient computational algorithms, analyze the performance of the methods for noisy data and apply them to microarray data classification.This proposed work will provide effective solutions to data biclustering, which is well known to be an important, but difficult and inherently intractable problem in general. The research results will also have many useful scientific and commercial applications, such as medical diagnosis, genetic network inference and information processing and retrieval.