A statistical critique of normalization methods in basic science research

Abstract

In basic science experiments, there is a tendency to utilize group-specific control measures in the creation of normalized quantities for comparison across groups. The desire is to compare relative knockout and wild-type gene expression measures in order to make claims about the effectiveness of a given treatment. These relative quantities represent a way of normalizing treatment data to a corresponding control. In this thesis, we explore two such types of normalization seen in research: fold change and comparative CT analysis using Reverse Transcriptase Polymerase Chain Reaction (RT-PCR) data. Within each method, we detail the derivations of relative quantities, walk through the statistical considerations being made when analyzing, and examine patterns seen in various simulation settings. We also make methodological recommendations to researchers such that they can maintain statistical validity in future analyses. In current research, statistical analysis of relative quantity data is often limited to an evaluation of the p-value from a two-sample Student’s t-test for difference in means. Through simulation study, we have identified Type I error rate inflation in many normalization settings due to violations of statistical test assumptions. These findings raise concern about the validity of conclusions made in previous research and highlight the importance of using statistically sound analytic techniques in the future.