Latent growth curve modeling of incomplete timecourse data in microarray gene expression studies

The microarray-based gene expression time-course experiments represent an important research design in biomedical studies. Current methods for analyzing microarray time-course data ignore individual differences in treatment responses or time-course patterns and have difficulties in handling missing values. We introduce the latent growth curve models popular in use in longitudinal epidemiology studies to analyze microarray time-course data with incomplete observations. The models have been applied to human in vivo irritated epidermis data with missing observations to investigate time-dependent global transcriptional responses to a chemical irritant. Our strategic data analysis identified significant time-course genes and selected the best fitting model for each of the genes on the array without pre-defining any parametric form. Various time-course patterns are then recognized from the significant genes using clustering methods including hierarchical clustering and self-organising map. Very high correlation between baseline expression and time-course response has been detected for many of the significant genes. The latent linear growth curve model can be applied to microarray-based gene expression time-course data with incomplete observations to find genes exhibiting various time-course responses. This method also estimates covariance between baseline gene activity and time-course response, providing useful additional information ignored in current gene expression analysis.