I have gene expression data from different conditions from different studies. Instead of using the actual TPM values for Pearson Correlation coefficient (PCC) calculation, I have decided to use Fold change values from different studies to eliminate biases from different studies. My question is whether using these raw fold change values for identifying co-expressed genes is a correct way to do it or should perform quantile normalization on these fold change values before using them for PCC calculation? (Note: Distribution of fold change values in different studies is quite different)
There are several methods to account for the differences between studies. As far as I know most of these methods use the raw probe intensities (microarrays) or gene counts (sequencing). The differences between studies or samples in the same studies that were processed in different groups are known as batch effects. You can look up the appropriate methods for removing batch effects in the specific data type you are handeling. Normalization would only be useful in removing the variance among samples in the same batch or after removing the patch effects.
I am not sure how were you planning to use fold-change to calculate gene co-expression to start with! The aim of co-expression analysis is to find the correlations between pairs of genes in a dataset or sets of data using a co-expression measure such as PCC. You don't need to calculate fold-change at all. The difference in gene expression between two conditions in a dataset is not the concern here. I will try to explain with an example
Expression matrix:
c1 c2 c3 t1 t2 t3
g1 1 1 1 2 2 2
g2 1 1 1 5 5 5
g3 6 6 6 3 3 3
The fold-change (t_vs_c) would be something like:
g1 = 2/1, g2 = 5/1, and g3 = 3/6
The co-expression would be something like:
g1_g2 = 1, g1_g3 = -1, g2_g3 = -1
There is however something called the differential co-expression which is the change in the correlation of a pair fo genes between two conditions. First, the correlation is calculated between a pair fo genes in each condition separately. Then the correlations are compared between conditons.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis