Dear all,
I seek some help with normalization strategy for my experiment. I have described the experimental design below.
Tissue gathered one wild type and 8 mutants - so genotype may be considered 'factor #1'.
For each of these genotypes, 4 time points have been used for library generation - so time may be considered 'factor #2'
The reason why I do not clump libraries across ALL genotypes is because I think it is like comparing apples to oranges to bananas to peaches....
Since the statistical validity of RNA-Seq comparisons allows for only a small fraction of DE genes in a background of largely unchanged gene expression, I think that comparing different mutants and wild type will violate this assumption. Do you agree? Or do you think I'd have to empirically prove this theoretical prediction before I conclude its a 2-factor experiment?
Now, IF you agree that the experiment is indeed a 2-factor one, then how do I go about
1. choosing the reference column / library for TMM normalization - should this TMM normalization be performed for one genotype at a time? I am leaning towards TMM normalization for each genotype separately.
2. since each library is quadruplicated, does edgeR allow independently replicated reference libraries to be ALL used for normalization against, or can I use just one of the 4 reference libraries at any one time (which would defeat the purpose of replication)?
For the purpose of creating a ref library, I a am thinking of making an RLE-based (geometric mean) pseudo-library from the 4 lib reps for the reference 'conditions'.
However, before calculating the pseudo-ref library, for the 4 ref libs I am first considering removing genes / rows where any expression falls beyond +/- 3 SDs. Your opinions on this?
To provide a context for my questions, the final goals of my research are to:
a. cluster and identify co-expressed genes within a genotype, and
b. identify genes with variant expression patterns across genotypes
I seek some help with normalization strategy for my experiment. I have described the experimental design below.
Tissue gathered one wild type and 8 mutants - so genotype may be considered 'factor #1'.
For each of these genotypes, 4 time points have been used for library generation - so time may be considered 'factor #2'
The reason why I do not clump libraries across ALL genotypes is because I think it is like comparing apples to oranges to bananas to peaches....
Since the statistical validity of RNA-Seq comparisons allows for only a small fraction of DE genes in a background of largely unchanged gene expression, I think that comparing different mutants and wild type will violate this assumption. Do you agree? Or do you think I'd have to empirically prove this theoretical prediction before I conclude its a 2-factor experiment?
Now, IF you agree that the experiment is indeed a 2-factor one, then how do I go about
1. choosing the reference column / library for TMM normalization - should this TMM normalization be performed for one genotype at a time? I am leaning towards TMM normalization for each genotype separately.
2. since each library is quadruplicated, does edgeR allow independently replicated reference libraries to be ALL used for normalization against, or can I use just one of the 4 reference libraries at any one time (which would defeat the purpose of replication)?
For the purpose of creating a ref library, I a am thinking of making an RLE-based (geometric mean) pseudo-library from the 4 lib reps for the reference 'conditions'.
However, before calculating the pseudo-ref library, for the 4 ref libs I am first considering removing genes / rows where any expression falls beyond +/- 3 SDs. Your opinions on this?
To provide a context for my questions, the final goals of my research are to:
a. cluster and identify co-expressed genes within a genotype, and
b. identify genes with variant expression patterns across genotypes