A Hidden Markov Model Approach for Prediction of Genomic Alterations from Gene Expression Profiling

摘要

The mRNA transcript changes detected by Gene Expression Profiling (GEP) have been found to be correlated with corresponding DNA copy number variations detected by Comparative Genomic Hybridization (CGH). This correlation, together with the availability of genome-wide, high-density GEP arrays, supports that it is possible to predict genomic alterations from GEP data in tumors. In this paper, we proposed a hidden Markov model-based CGH predictor, HMM_CGH, which was trained in the light of the paired experimental GEP and CGH data on a sufficient number of cases, and then applied to new cases for the prediction of chromosomal gains and losses from their GEP data. The HMM_CGH predictor, taking advantage of the rich GEP data already available to derive genomic alterations, could enhance the detection of genetic abnormalities in tumors. The results from the analysis of lymphoid malignancies validated the model with 80% sensitivity, 90% specificity and 90% accuracy in predicting both gains and losses.