PDX Samples

[gabi-ryan]

Hi, just looking for some advice -

I am looking at processing tumor-only PDX samples from targeted sequencing. I noticed this workflow on Cavatica and you mention using the Xenome tool for filtering out mouse reads for PDX samples. I can't find anymore information on this - do you have this implemented in the workflow or do you recommend filtering beforehand? I have tried running Xenome myself however it takes a while to index & classify. I'd love to hear If you have any advice / further information regarding this type of data.

Thanks,
Gabi

[migbro]

Hi @gabi-ryan , thanks for your interest in this workflow. We haven't yet baked it in as an optional tool to run, but in general, It'd say it'd go like this:

1. First run xenome index. This can be reused for other PDX samples, so you only need to run this once. It's been a while, but indexing using the human hg38 ref fasta as graft, and mouse mm10 as host took a little over 2 hours, with 16 cores and 128 GB memory.
2. Next, for each PDX sample, you'd run xenome classify. I don't have a good sense of how long it'd take to run on WGS, for RNA it was less than two hours.
3. Use the graft_fastqs output to run the tumor-only workflow. This should be the closest you'll get to removing any mouse contamination from the human sample before calling.
   Conversely, if you're pretty confident that the PDX excision has little to no mouse contaminant, you could roll the dice and use the reads as-is for the tumor-only workflow. Hope this helps!

[migbro]

Hi @gabi-ryan , I was able to find an example where we used Xenome Classify on WGS, and it was quite a long run, basically 24 hours. Unfortunately, I don't quite have the bandwidth right this moment, but I do have a suggestion for improvement. I do beliee if you can chunk your fastqs, into at least 4 pieces say, we can get that down to 6 hours if you run each quarter as it's own task, then concat the results after. Otherwise, I'd probably turn this into a scatter gather workflow that'd do this for you.

[gabi-ryan]

Hi @migbro thank you for the advice. I was able to index the reference, but I found that with classify it seemed to hang for more than 2 days - I think this is a known issue with Xenome unfortunately, and it hasn't been updated in about 8 years. I was able to run bbsplit from BBMap instead, which was fairly quick but requires a lot of memory, I also have yet to assess the quality of this method. I might continue to look into Xenome and split it into chunks as you have suggested and compare the results with bbsplit. Cheers.