Analysis of qPCR data

Question

[simonhb1990]

Hi all,

I'm not familiar with the field of real time PCR and how to deal with the data. However, I need to reproduce the result (Fig. 1A) of the paper as follow: http://www.ncbi.nlm.nih.gov/pubmed/20412781

Basically, it is a hierarchical clustering on single cell qPCR data.

I follow the methods part of the paper for the Table S4 in the paper's SI:

All Ct values (Tables S4 and S5) obtained from the BioMark System were converted into relative expression levels by subtracting the values from the assumed baseline value of 28. Cells with low or absent endogenous control gene expression levels were removed from analysis (~10%). The resulting values were at times normalized to the endogenous control by subtracting, for each cell, the average of its Actb and Gapdh expression levels. As the Ct scale is logarithmic (a difference of one Ct corresponds to a doubling of measured transcript), a subtraction of the average of two genes on this scale corresponds to taking the geometric mean on a linear scale. Data shown in Figures 1A-1C, 2A, 3A, 3B, 4A, 5A, and 5D have been normalized against endogenous controls.

After subtracting from 28 and normalized the data by endogenous genes, the resulting values are mostly negative.

I think maybe I need to further normalize the data by calculating the zscore over each gene and also over each sample, but I'm not sure if these normalization make sense in this field or nor?

Really appreciate if anyone could have a look at the data and let me know how to reproduce it.

Regards,
Simon

Answer 1

[Devon Ryan]

Their wording is rather poor, but you can ignore the "subtract 28 from the Ct value" part. Just average the endogenous controls and subtract the resulting averaged Ct value from that of your gene of interest and you have normalized relative values (on the log2 scale). If you happen to get a negative value then that's fine, it just means that whatever gene you're looking at is highly expressed.

[simonhb1990]

Thanks very much for your reply. For the normalized relative values, do I need to normalize them (center the mean and divide by the Std) again? I think for microarray data, people usually do it.

[Devon Ryan]

Heatmap functions typically offer to center and scale each row, so yes that would typically be done as well. This is separate from how the data was actually generated.

[simonhb1990]

Thank Devon, I think I'm clear with it.