Multilevel modeling in human microarray time course gene expression data

In this study different numbers of models were learned over a single dataset of human microarray time-course gene expression data. The idea of running this extensive broad of modeling was lack of true knowledge of variance covariance of error structure among gene expression vales due time. The results indicated that it is difficult to find the right and sufficiently accurate variance covariance among gene expression data which defines the expression values the best. Also, the results indicated that there is less small negative correlation between intercept and slop among expression values within genes. This generally indicates those genes which get started with high expression values, likely decreases their expression across times. The analysis suggested that multilevel modeling can likely pinpoint to differentially expressed genes across whole dataset. As a side it concluded that by using multilevel modeling, genes which show periodic activity across time can be winnowed down too.