Autoscaling and mean centering

[MultiD Support]

Hello,
Shall I autoscale or mean center my data for analysis?

[MultiD Support]

Mean centering removes the dependence of the average expression of the genes, mut retains the magnitude of its fluctuations. Autoscaling removes also the magnitude of fluctuations.

Lets assume that you compare the efficiency of two drugs by measuring the response of a target gene as function of time after administratinf the drugs. The two drugs give rise to the the same temporal response but with twice the amplitude of the first drug. Most likely you would conclude that the first drug is more efficient.

If you instead treat with only one drug and measure the temporal response of two genes, and they show the same same response but with gene 1 having twice the amplitude of gene 2. Most likely you would conclude that the two genes show the same behavior.

The first conclusion you reach if you analyze mean centered data, while the second conclusion you reach if you analyze autoscaled data.

Hence, ruel of thumb is to use mean centered data when classifying samples and autoscaled data when classifying genes.

Mikael