Hi All,
I'm currently trying to do differential expression analysis comparing expression differences at orthologous loci between two closely related taxa that have been aligned to separate references. The issue is that I am hoping to directly compare two count values (# of reads uniquely aligned) that were obtained from alignment to separate reference sequences of different lengths.
i.e.
-sample 1_taxa1_contigX_ortho1_contiglength1257 has a count of 1200
-sample 2_taxa2_contigY_ortho1_contiglength1989 has a count of 57
can I do differential expression analysis to see if these two expression levels are constitutively different?
So far there are two separate issues:
1-Normalizing for sequencing depth-
a-Most DEG analysis programs (e.g. DESeq, edgeR, etc.) do a within experiment normalization for sequencing depth, however, I have two separate references with a subset of orthologous loci so normalizing different counts at the subset orthologs only would ignore the counts at non-orthologs. This doesn't seem appropriate for other reasons as well.
2-Accounting for differences in contig length-
a-Since contigX for ortholog1 from taxa1 is a different length than contigY for ortholog1 from taxa2, there could be some length bias in alignment efficiency that could mess up the differential expression analysis.
My proposed solution would be to resurrect RPKM normalization for this particular case because it independently normalizes each sample based on total number of reads and the different contig lengths. I would do this:
1- RPKM normalization for all reference contigs for taxa1 and separately for all reference contigs for taxa2
2- Pull out the counts for the ortholog subset
3- Combine the data from each taxa at the orthologs
4- Do some type of differential expression analysis comparing counts from taxa1 at ortholog1 to taxa2 at ortholog1
However, I am not aware of any DE package that will take RPKM normalized data. I've managed to trick DESeq into doing it by using a sizeFactor of 1 after RPKMing my data and rounding to integers but this seems less than rigorous and/or true to the statistical framework.
Any suggestions for other methods to account for the different sequencing depths and different contig lengths, as well as possible statistical frameworks to analyze custom normalized data would be much appreciated.
This is for a microbe system where we are trying to investigate expression differences among different genotypic populations of organisms that are divergent enough to warrant separate reference assemblies but have a well matched subset of orthologous loci.
Any assistance or links to other posts that might be helpful would be much appreciated.
Many thanks,
dan
I'm currently trying to do differential expression analysis comparing expression differences at orthologous loci between two closely related taxa that have been aligned to separate references. The issue is that I am hoping to directly compare two count values (# of reads uniquely aligned) that were obtained from alignment to separate reference sequences of different lengths.
i.e.
-sample 1_taxa1_contigX_ortho1_contiglength1257 has a count of 1200
-sample 2_taxa2_contigY_ortho1_contiglength1989 has a count of 57
can I do differential expression analysis to see if these two expression levels are constitutively different?
So far there are two separate issues:
1-Normalizing for sequencing depth-
a-Most DEG analysis programs (e.g. DESeq, edgeR, etc.) do a within experiment normalization for sequencing depth, however, I have two separate references with a subset of orthologous loci so normalizing different counts at the subset orthologs only would ignore the counts at non-orthologs. This doesn't seem appropriate for other reasons as well.
2-Accounting for differences in contig length-
a-Since contigX for ortholog1 from taxa1 is a different length than contigY for ortholog1 from taxa2, there could be some length bias in alignment efficiency that could mess up the differential expression analysis.
My proposed solution would be to resurrect RPKM normalization for this particular case because it independently normalizes each sample based on total number of reads and the different contig lengths. I would do this:
1- RPKM normalization for all reference contigs for taxa1 and separately for all reference contigs for taxa2
2- Pull out the counts for the ortholog subset
3- Combine the data from each taxa at the orthologs
4- Do some type of differential expression analysis comparing counts from taxa1 at ortholog1 to taxa2 at ortholog1
However, I am not aware of any DE package that will take RPKM normalized data. I've managed to trick DESeq into doing it by using a sizeFactor of 1 after RPKMing my data and rounding to integers but this seems less than rigorous and/or true to the statistical framework.
Any suggestions for other methods to account for the different sequencing depths and different contig lengths, as well as possible statistical frameworks to analyze custom normalized data would be much appreciated.
This is for a microbe system where we are trying to investigate expression differences among different genotypic populations of organisms that are divergent enough to warrant separate reference assemblies but have a well matched subset of orthologous loci.
Any assistance or links to other posts that might be helpful would be much appreciated.
Many thanks,
dan