A Unified Statistical Strategy for Normalization and Hypothesis Testing
One of the difficult aspects of dealing with genomic data is the presence of heterogeneity between chip experiments and the need for normalization. As mentioned above in the case of microarray data, gene expression values within a chip are measured on a relative scale, i.e., values within chips are comparable but they are not directly comparable across chips.
To facilitate any meaningful analysis, one has to normalize data so that expression values across different chips are comparable. We have developed a regression strategy that includes these variations that we have successfully applied to a variety of statistical models. We have successfully applied this Heterogeneous Regression Framework (HRF) to a variety of gene expression studies.
By using the techniques of regression (linear models, logistic regression, Cox models and generalized linear models) the HRF makes it possible to assess the results of gene expression studies using tools that are commonly used in biostatistical practice. We believe that this is a major strength of our work. Furthermore, because we use models that are part of common statistical practice, we can address issues of False Discovery Rate, type I and type II errors.