Quick question about analysis of log2 RMA data

Question

[mforde84]

Say I have some probe level and RMA normalized array data. The data is supposedly log2 normalized already. How do I calculate the relative difference in intensity between two probes?

If I understand correctly, it should just be the 2^residual.

For example

probe1 = 1
probe2 = 2

2^(2-1) = probe2 has 2x as much signal as probe1

However for values less than 1, I'm not sure this makes sense:

probe1 = 0.25
probe2 = 0.5
2^(0.5-0.25) = probe2 has 1.19x as much signal as probe2?

Shouldn't it be 2x as well in this instance?

[mastal511]

If your values are already log2 of the intensities, then you get the difference by subtracting the log2 values, because log(A/B) = log(A) - log(B). This is equivalent to dividing the initial (base 10) intensities, A/B, so essentially it gives you a fold-change.

[mforde84]

If I understand correctly the resulting fold change is still in log2, correct?

So 0.5 is not actually double the intensity of 0.25, where 2 = 0.5 / 0.25. But instead 1.19x because 2^(0.5-0.25).

And 2^( (0.5-0.25) / 0.25) doesn't make sense because log2 isn't linear.

[mastal511]

Yes, you're right, you get a log2FoldChange rather than a Fold-Change. 0.5 is not double the intensity of 0.25 because those were the log2 values of the intensities.