4

There are wet methods: Patient-derived models: Patient Derived Xenograft (PDX) and Patient Derived Organoids (PDO) to reflect tumor biology.

Are there any databases or computational tools that use the outcomes from PDO/PDX experiments to create a predictive computational model for cancer or other diseases?

0x90
  • 1,437
  • 9
  • 18
  • 1
    What do you mean by outcomes of PDO/PDX experiments? RNA-seq of the tumor? Clinical variables along time for PDX experiments? Interactions between cell lines in the PDO experiments? What do you want to predict: mortality, growth, some clinical variable value, metastasis...? AFAIK they are used to learn about the tumors, not to predict anything. – llrs Jun 28 '17 at 09:02
  • @Llopis I think that all what you have just listed could use to build a predictive model. – 0x90 Jun 28 '17 at 09:07
  • But what do you want to predict ? And more important, would that be a good model for what? – llrs Jun 28 '17 at 09:08
  • @Llopis by prediction I mean to build a general model to predict if a patient has cancer and which medicine would be the most suitable for him. – 0x90 Jun 28 '17 at 09:11
  • You can't predict if a patient has cancer from a cancer model! If you want to predict which medicine for a given patient's tumor would be the most suitable for him you need to test (in silico or not) all the medicines you think they have a change. (Also you would have already removed the tumor or part of it for both methods, so presumably the patient would no longer have it given the current techniques). In my opinion, this question is too broad, if you have a concrete example or a paper doing something similar it could be answerable but as it is, is too broad and lacks many details. – llrs Jun 28 '17 at 09:15
  • @Llopis saying too broad is a nice way of saying not feasible? – 0x90 Jun 28 '17 at 09:17
  • It is feasible, but this question is not answerable. – llrs Jun 28 '17 at 09:19
  • @Llopis https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4323263/ – 0x90 Jun 28 '17 at 09:36
  • 1
    Let us continue this discussion in chat. – llrs Jun 28 '17 at 09:39
  • @Llopis I am experiencing some difficulties to login. Please accept my apologies. I am looking for something similar to pan-cancer, other cancer types and Alzheimer's. moreover as you mentioned that's only PDX. – 0x90 Jun 28 '17 at 09:49
  • It is Ok. Do your on due diligence and search in the literature for it, when (if) you find yourself stuck ask for the concrete problem you are in. But avoid asking a broad general question. – llrs Jun 28 '17 at 10:14
  • @Llopis thanks, I still don't think it's too broad, but appreciate your feedback and will try to update it. – 0x90 Jun 28 '17 at 10:27
  • this question asks 2 quite unrelated questions, it would be best to separate these into 2 questions. – nsheff Jun 30 '17 at 19:57
  • @nsheff do you mean the first and the secondly? They are closely related. – 0x90 Jun 30 '17 at 19:59
  • @0x90 There's no good mechanism to split a question other than to edit it and post a new one. Having said that, your first question is appropriate for the biology SE and only the second one for here. – Devon Ryan Jul 01 '17 at 06:07

1 Answers1

1

Potential pitfall!

I'm not sure about predictive models, but you need to be aware of a potential pitfall in blindly aligning PDX or PDO based sequencing data without first removing contaminating host organism reads, as otherwise these will lead to a lot of false positive variants caused by miss alignment. In my experience even a small mount of host material can lead to a ten fold increase in called variants due to miss aligned host reads looking like true variants. I'd recommend using Xenome, source.

Note Xenome was designed for DNA sequencing it's not splice-aware, so for RNA-Seq there is no optimal solution other than removing reads which align to the host organisms genome. Although the issue here is that in conjunction with a splice-aware read aligner synteny between the host and grafted genomes might create some interesting problems. However for RNA-Seq provided FACS or similar confirms low-levels of host contamination I expect levels of expression will not be badly affected. Although this really needs investigating.

Finally and rather annoyingly Xenome produces none-standard FASTQ so you'll need to fix it's output with:

awk '{if (NR % 4 == 1) print \"@\"$0; else if (NR % 4 == 3) print \"+\"$0; else print $0 }' as reported here on seqanswers.

Matt Bashton
  • 1,069
  • 6
  • 16