Smoothing Spline Mixed Effects Modeling of Multifactorial Gene Expression Profiles

The analysis of time-course microarray data is challenging because of the wide range of gene expression patterns, multiple sources of variation, and dependency of the measurements. The performance of smoothing spline mixed effects models to describe time-dependent gene expression trajectories, technical and experimental sources of variation were studied. The comparison of spline, polynomial ANCOVA and ANOVA models allowed us to characterize the balance between model simplicity and adequacy to describe microarray element expression trajectories. Gene expression measurements at honey bee behavioral maturation time points with, with deviations from the overall pattern across honey bee races and host colonies were analyzed. Complementary criteria were used to compare model adequacy including, visual comparison of expression trajectories, number of microarray elements with significant time-dependent terms and consistent time-dependent trends across models. Spline models were favored for the vast majority of the microarray elements when model fit and model simplicity criteria were considered.