RNA-seq analysis of two species

Question

[afli]

Dear all,

I have RNA-seq samples of two related species with similar genomes. The quality of genomes are good, is there any ways to map and compare the differential genes between them?

As far as I know, kallisto can calculate expression levels for each transcripts in each species, but this would require me to do some blat jobs to find the homologous genes between these two species and then do comparation. I just wonder if there is any other better ways? Thank you!

Aifu.

[corend]

Hi, Similar question here

[afli]

Thank you. I see the post, in your case, you have one genome that is of low quality. This is different from mine, the two genome are both of high quality, and they are similar. If I align the two sample reads to one reference genome, the reference genome itself is some different from the two genome. Is is possible to do something more precisely?

[corend]

I am not an expert, I don't want to tell you a wrong answer, but in your case maybe I would check the genomes similarity by mapping reads from species 2 to species 1 (or/and species 1 to species 2) and see the overall alignment rate. And if it is good enough for what you want to do, use this genome as a reference (this is still a non expert opinion).

[afli]

OK, thank you for your advice.

[Martombo]

Depending on what species you're working on, you could find ortholog information in ensembl. I think that would be a much better approach than mapping the RNA-seq of the two species to the genome of one of them.

[afli]

All right, thank you~

Answer 1

[Charles Warden]

If they are closely related genomes, it might be worth considering an RNA-Seq assembly from a related genomic alignment (using an algorithm like cufflinks)? If so, it would probably be good to have multiple samples (where I think the cuffmerge assembly is probably better than any of the individual sample assemblies).

If you have genomic sequence, you could also try using a program like MAKER (where you can provide RNA-Seq data in the annotation process). However, that may take a while for a whole genome sequence (particularly if it is a vertebrate genome).

Otherwise, I sometimes use Oases (or maybe even Velvet contigs) for RNA-Seq de novo assembly, but I would usually be assembling partial transcript sequences (most likely, for a fraction of the total transcripts in any sample). In other words, I would have some concerns about using that assembly for quantification. However, if there is a set of RefSeq sequences for your organism (I would guess from ESTs, and other sources), that may be the best option for transcript quantification.

[afli]

Thank you Warden for your detailed explanation.

[corend]

If they are closely related genomes, it might be worth considering an RNA-Seq assembly from a related genomic alignment (using an algorithm like cufflinks)? If so, it would probably be good to have multiple samples (where I think the cuffmerge assembly is probably better than any of the individual sample assemblies).

This "old" tuxedo (Tophat2 Cufflinks) should not be used anymore, the authors recommend now the New tuxedo pipeline using Hisat2 StringTie Ballgown.

[Charles Warden]

I've definitely found examples where it was better to use TopHat2 over HISAT2 (and I also perfer STAR over HISAT).

I suspect that there are probably also some advantages to using cufflinks. However, to be honest, besides emphasizing the value in using cuffmerge (which I think adds more robustness than the assembly for any individual sample), I don't use cufflinks very often (and there could be other factors that I am not considering). So, I would guess is the --merge mode in StringTie could be important, but it might be better for somebody else to share their experience with that program.