RNA-seq w/ combined mouse+human tissue: public data & methods?

[turnersd]

I'm proposing to do some RNA-seq analysis of human cancer tissue transplanted into a mouse (mouse tissue stroma around the tumor), and look at expression in both human and mouse transcripts. I'm guessing I'll have around 90% human, 10% mouse tissue.

My primary question: is there any public data (GEO, SRA) that has samples like this? I'd like to use public data to assess feasibility and test a few strategies for mapping this data. Which leads to question #2:

Secondary question: what's the best way to map this data? I see a few options: (1) map all reads to human, remainder to mouse (or vice versa), (2) map all reads to human, then all reads to mouse, or (3) map all reads to concatenated reference index, eliminating multimappers. I'm thinking #3 would be best, but requires some extra legwork creating combined fasta files, combined indexes, then disentangling reads that align to human vs mouse in the downstream alignment.

[dstorey]

I can't help you with 1, but I have a similar project of mixed transcripts from different species.

While the genomes were no where near as large , we simply went with concatenating them together and doing the search I doubt you'll have any memory issues if you're on a modern machine (>8Gb RAM).

We used Bowtie and ramped up the number of time we re-seeded the search. (Well above the very-sensitive pre-sets that are recommended) We also made the decision not to separate the reads down stream until after we got normalized counts but I don't think separating multi-mapped reads out will be as difficult as you think.

[turnersd]

Thanks dstorey. I also recently came across this method for partitioning reads to their respective species before mapping. Might give it a try.

http://www.ncbi.nlm.nih.gov/pubmed/22689758

[dstorey]

Not to be a smart ass but will this tool increase mapping accuracy or decrease computational complexity in a meaningful way? Otherwise a simple grep would help you sort mapped reads based on where they mapped.

[Simon Anders]

I'd agree with the OP that his method #3 sounds best. We did this recently in another experiment involving mixed material from two species and it worked very well. And concatenating a few FASTA files is not reallt that much effort.

One caveat, though: Some aligners seem to have issues if the reference has more then 2^32 bp (i.e., 4 Gb), and human alone is already 3 Gb. We used GSNAP, because it can handle large references, at least since recently.

[Jeremy]

I would also go with number 3. Most mappers I have used will map a read to the best fit that it finds meaning:
for 1) that whichever species you chose to map first would get reads from the other species mapping for conserved regions;
and for 2) there will be many cases where a read maps to both mouse and human and will require additional work to determine which is the best map

[bioBob]

Hi Stephen.

We are about 1 month away from having this type of data for human-mouse and dog-human transplants so the post is timely for us. We were planning on going down your option 3 as well, but I do like the idea presented in the Xenome paper.

[turnersd]

Hi Stephen.

We are about 1 month away from having this type of data for human-mouse and dog-human transplants so the post is timely for us. We were planning on going down your option 3 as well, but I do like the idea presented in the Xenome paper.

Bob (is that you Settlage?) - I'd love to see what you think if you try both approaches.

[bioBob]

Yes, we can talk after we get some data. Hopefully before ABRF.

[scottyler89]

I realize that I'm probably coming into this conversation a bit late. But - If it were me - I'd probably try to separate them biologically before sequencing. I would use something like SCID mice with ROSA-tomato, or another reporter, so that you could use RNA-later and separate by flow.

I agree with everyone here that approach #3 would work best, but there could be interesting data within those genes. Just a thought - if you hadn't yet started and had access to reporter SCID mice.