As discussed in the Objectives of image processing section, even under the same condition cells can be at different states with different expression levels due to various factors, such as cell cycle status, drug efficiency, heterogeneity of the cell population, etc. Hence, rather than obtain an average expression level, one should aggregate the cell-wise expression levels from all duplicate wells to obtain the expression profile across the whole population. If, for a given gene, the expression state of a cell at a certain time point is viewed as a random distribution, then the expression profile is a density function depending on the cell line, gene, external stimulant, and time. Although, histograms are widely used to visualize the density function, it is cumbersome to put different profiles of different time points and conditions in one plot for comparison, which is critical to the time-course case-control study. Thus, we use a one-dimension kernel density estimation with Gaussian kernel to obtain a density function, which has the properties of being nonnegative and having integral one.22,23 The smoothing parameter is selected by the Sheather-Jones method24 implemented in the Matlab toolbox as discussed by Marron in 25. Figure 5(a) shows expression profiles in the form of density functions obtained for the cell population imaged in Fig. 1. The profiles of all 48 time points are shown. The lines are coded by grey-scale to indicate time. The two time points corresponding to the two images shown in Fig. 1 are shown in bold lines, with the before-the-drug time point in black, and 43-h time point in light-grey. As a comparison, Fig. 5(b) shows the expression profiles of the un-drugged population of the same reporter and cell line. Note that in this example, since both plots are based on a single imaging site, the density curves are actually not as smooth as pooled results. The -axis is the transcriptional activity level measured by the total fluorescent intensity in scale. The figure clearly shows that the transcription level of target gene MKI67 is relatively high at the beginning of the experiment, with the peak at around 217 (arbitrary camera intensity units). At the end of the experiment, the transcription level of MKI67 has been greatly reduced to around 214, which is a decrease of roughly 8-fold, and is close to the fluorescence level that a cell without GFP would exhibit. The transcription profile also shows that there might be a small portion of cells that remain in a highly proliferative state as their MKI67 levels show no sign of decrease.