Hi all,
I have to do differential gene expression on my RNA-seq data (novel organism, reference transcriptome) and the more I read about how to do it the more confused I get. A colleague of mine has looked into Partek for his NGS differential expression work and it was suggested to me to do the same to maybe circumvent running all the programs individually and trying to figure it all out on my own. Now Partek seems to use RPKM for their "mRNA quantification step", which I have read is not so good for paired-end data as it counts the reads and can bias the counts in certain situations (when I have pairs as well as only one mate of the pair aligning). When I asked if I could at least switch to FPKM for normalisation they told me that if I use paired-end reads with strict paired-end compatibility option, RPKM equals FPKM.
Now I'm not a statistician, but wouldn't that bias as well? Or is it general practice to just use the reads that align as pairs, and not include the reads where only one read of the two aligns to the reference? I have seen bowtie2's default is to align pairs and also one mate of the pair if it can only find an alignment for one of the two reads. There is a command that would allow me to force bowtie2 to only look for paired alignment and nothing else. I didn't read anything similar in the bowtie manual and I have only ever used bowtie in a pipeline, so never saw the output like I did with bowtie2. Does bowtie work the same as bowtie2, or does bowtie only look for paired alignment and nothing else straight from the beginning? What would be the best strategy for differential gene expression work, to align only pairs or to try and align as much as I can, even if it means aligning pairs and one mate of the pairs (ie unpaired alignment).
Thank you!
I have to do differential gene expression on my RNA-seq data (novel organism, reference transcriptome) and the more I read about how to do it the more confused I get. A colleague of mine has looked into Partek for his NGS differential expression work and it was suggested to me to do the same to maybe circumvent running all the programs individually and trying to figure it all out on my own. Now Partek seems to use RPKM for their "mRNA quantification step", which I have read is not so good for paired-end data as it counts the reads and can bias the counts in certain situations (when I have pairs as well as only one mate of the pair aligning). When I asked if I could at least switch to FPKM for normalisation they told me that if I use paired-end reads with strict paired-end compatibility option, RPKM equals FPKM.
Now I'm not a statistician, but wouldn't that bias as well? Or is it general practice to just use the reads that align as pairs, and not include the reads where only one read of the two aligns to the reference? I have seen bowtie2's default is to align pairs and also one mate of the pair if it can only find an alignment for one of the two reads. There is a command that would allow me to force bowtie2 to only look for paired alignment and nothing else. I didn't read anything similar in the bowtie manual and I have only ever used bowtie in a pipeline, so never saw the output like I did with bowtie2. Does bowtie work the same as bowtie2, or does bowtie only look for paired alignment and nothing else straight from the beginning? What would be the best strategy for differential gene expression work, to align only pairs or to try and align as much as I can, even if it means aligning pairs and one mate of the pairs (ie unpaired alignment).
Thank you!