Microarray Gene Expression Analysis

Although considered modern and state of the art, available software (commercial and/or free) is based on statistical methodology of microarray data analysis that is far from satisfactory. The currently practiced methods of significance testing in microarray gene expression profiling are highly unstable and their power tends to be extremely low. These undesirable properties reside in the nature of multiple testing procedures and the well-documented extremely strong and long-ranged correlations between gene expression levels. Resorting to normalization procedures does not provide a solution to the problem because of their distorting effects on the true expression signals. Such effects are especially pronounced in large sample studies where control of type 1 errors may be entirely lost. While the commonly used normalization methods may be beneficial under certain conditions, there is no way of testing for the presence of such conditions in a specific data set.

The Division of Computational Biology (DCB) within the Department of Biostatistics and Computational Biology has systematically studied numerous pitfalls in microarray data analysis for more than 5 years. All major methods of gene expression profiling have been rigorously tested by resampling from large data sets and none of these methods has passed this test. This led us to conclude that the currently available commercial software is not statistically sound and should not be used for the analysis of microarray data. The only method that meets the requirements of statistical rigor and by far outperforms all other methods was recently developed by members of the DCB. This method exploits a specific feature of gene expression data, termed the delta-sequence, to significantly improve the most basic performance indicators. ^1,2

We offer this newly developed method to the scientific community at the University of Rochester. The method built on the concept of delta-sequence can be used in conjunction with the conventional multiple testing procedures that control the family-wise or per-family error rates and the false discovery rate, as well as with the empirical Bayes methodology. We also offer other methods for microarray data analysis such as a multivariate extension of the gene set enrichment analysis, multivariate search for differentially expressed gene combinations, and spectral analysis of hidden periodicities in chromosomal gene expression patterns. Our rigorously tested methods are statistically sound and offer the precise results necessary to the work of our colleagues.

1. Klebanov, L. and Yakovlev, A. Diverse correlation structures in microarray gene expression data and their utility in improving statistical inference, Annals of Applied Statistics, 2007, 1(2): 538-559.

2. Klebanov, L., Qiu, X., and Yakovlev, A.Y., 2007. Testing differential expression in non-overlapping gene pairs: A new perspective for the empirical Bayes method. Journal of Bioinformatics and Computational Biology, in press.

(Back to Consulting Service homepage)