Serum starvation: deconvolved expression
Microarray experiments measure the average RNA level of each gene in a population of cells,
and thus are most accurate when using a homogenous population of cells.
Partial synchronization causes a severe distortion of microarray results.
In order to overcome this problem we developed a computational approach that takes advantage of the
FACS data
collected at various time points along the experiment to deconvolve the expression data.
The deconvolution algorithm infers gene expression values for the ideal "average single cell";
it does this using the empirically observed distribution of cells and measured expression values
recorded at each time point. The algorithm is based on the assumption that
following release from arrest, each cell proceeds according to its own internal clock.
Some of the cells do not emerge from the arrested state, and the remaining cells proceed along the
cell cycle at their own rate which, assuming a normal distribution, can be inferred from the FACS data (see methods).
The inferred synchronization loss model can be applied to deconvolve the
measured expression data to generate "single cell" gene expression profiles.
Excel file containing deconvolved values for genes appearing on Whitfields' array
Excel file containing deconvolved values for genes that are not present on Whitfields' array