Hidden Markov model for defining genomic changes in lung cancer using gene expression data.

Abstract

The study of gene expression patterns in relationship to chromosomal position, the "transcriptome map," has become an area of active research and has revealed unexpected chromosomal regions within which gene expression levels are highly correlated. In cancer research, these regional changes in gene expression that may result from alterations at the chromosome level such as gene amplification or loss. To facilitate the search for such regions utilizing gene expression data, we have developed a hidden Markov model (HMM). Maximum penalized likelihood is used to estimate the parameters in the model. This method is applied to a lung cancer microarray experiment, including 86 human lung adenocarcinomas. Several regions identified through the HMM are consistent with known recurrent regions of amplification or deletion in this cancer. We further demonstrate the association of these abnormal expression regions with measures of disease status, such as tumor stage, differentiation, and survival. These findings suggest that genes in these regions may play a major role in the process of carcinogenesis of the lung. Our proposed method provides a valuable tool to accurately pinpoint regions of abnormal expression for further investigation.