I used to perform the normalization of microarray data from Affymatrix platform using RMA method. However the processed data tend to be smaller and the divergence between probes also become smaller. I know the reason is the logarithm. I wanna know the reduction of relative value of probes may affect the discovering the differentiated expressed genes. Or should I use the ratio between value before log?
When you take logs of expression values, you need to remember that the value obtained is now in "log base N". So for example if you used log base 2, a difference between values of 1 is actually a 2x difference, a difference between values of 2 = 4x difference and so on. So the smaller ratio between probes under different conditions when logged is not a "bad" thing; it's just a different way to express the numbers.
If there is differential expression, the test used to find it (e.g. p-values or something better, such as SAM) will take the magnitude of the expression values into account. So don't worry if it "looks small" to your eyes.