|
Abstract:
|
An important goal for genomics data analysis is to identify useful gene-gene interactions so as to discover potential functional groups of genes. Such analysis, however, is prone to false positive discoveries due to handling effects, which arise from technical variations encountered in the experimental process. Post-doc data adjustment via 'normalization' is typically used to smooth out the data and 'correct' for such handling effects. However, little is known on how normalization impacts the discovery of gene-gene interaction and whether a prior study designs based on statistical principles provides a better solution. We set out to assess the role of handling effects, data normalization, and study design on the discovery of gene-gene interaction, using a unique pair of microarray datasets on the same set of samples. One dataset was collected with uniform handling and balanced design for array-to-sample assignment to avoid confounding handling effects; the other dataset used non-uniform handling and exhibited handling effects. We report here some initial results from this empirical assessment.
|
