Robust cross-race gene expression analysis

Abstract

This paper develops a Bayesian network (BN) predictor to profile cross-race gene expression data. Cross-race studies face more data variability than single-lab studies. Our design handles this problem by using the BN framework. In addition, unlike existing methods that unrealistically assume independent genes, our BN approach can capture the dependencies among genes. Existing BN algorithms in biomedicine applications quantize data, leading to information loss; we adopt linear Gaussian model to keep the data intact, so our resulting model is more reliable. The application of our BN predictor to a lung adenocarcinoma study shows high prediction accuracy, and performance evaluation demonstrates our gene signature agreeable with those reported in the literature. Our tool has a promising potential in finding disease biomarkers common to multiple races.