Skip to content

[Bioc-devel] From Biostring matching to short read mapping

5 messages · Hervé Pagès, Wei Shi, Bhagwat, Aditya

Original
Bhagwat, Aditya
# 
Dear bioc-devel,

multicrispr<https://gitlab.gwdg.de/loosolab/software/multicrispr> provides functions for Crispr/Cas9 gRNA design (and is being prepared for BioC). One task involves finding all genomic (mis)matches of a 23-bp candidate Cas9 sequence. Currently this is done with `Biostrings::vcountPDict`, an approach that is successful, though not fast. An alternative would be to switch to short read mapping rather than (Bio)string matching, which involves a one-time indexing effort, but subsequent fast alignment.

`Rsubread::align` seems to be limited to max. 16 `nBestLocations`, whereas I know from vcountPDict that some Cas9 candidates have hundreds of genomic matches.

`QuasR::qAlign` (connecting to Bowtie) does not mention an upper limit on `maxHits`.

Feedback request...

Michael, would QuasR/(R)bowtie be a good approach to do this?
Wei, did I overlook a way to do this with Rsubread?
Herve, is there an elegant way to speed up vcountPDict (parallelize?)

Thankyou :)

Aditya
1 day later
Hervé Pagès
# 
Hi Aditya,

Should not be too hard to parallelize. With some gotchas: using one 
worker per chromosome (which is the easy way to go) wouldn't be optimal 
because of the size differences between the chromosomes. So a better 
approach is to try to give each worker the same amount of work by 
splitting the set of chromosomes in groups of more or less equal sizes.
The split can either preserve full chromosomes or break them in smaller 
pieces. The later will allow using a lot more workers than the former.
I'll try to come up with some code that I'll share here.

BTW the *PDict() family in Biostrings is for finding the matches of a 
collection of patterns. You say you want to find "all genomic 
(mis)matches of a 23-bp candidate Cas9 sequence". Any reason you're not 
using vmatchPattern() (or vcountPattern()) for that?

Cheers,
H.
On 11/7/19 02:11, Bhagwat, Aditya wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Bhagwat, Aditya
    

    

      
      #
    

  


  

      
Thankyou Michael, I got Rbowtie working, now functionalizing it for use within multicrispr. I noticed that in QuasR, you actually create a package with bowtie indices which you then use for future purposes. Interesting workflow, think I will make use of that functionality.

Thankyou Herve. Yes, parallellizing would speed up things. I use `vcountPDict` because I want to do the offtargetanalysis for a set of 23 bp cas9 sites. vcountPDict must be more efficient than looping, I thought, maybe this is only marginally so, I noticed there's an sapply underlying vcountPDict. Is there a BSgenome way to parallellize, like a parallel bsapply or so?

And Rsubread I concluded is really limited to only a small number of co-alignments, and so not suited for offtargetanalysis.

Cheers,

Aditya

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Wei Shi
    

    

      
      #
    

  


  

      
Hi Aditya,

Yes you are correct that Subread reports no more than 16 alignments per reads. One reason for this limitation is because subread detects indels in the read (Bowtie does not detect indels) and it has to set a limit on the number of candidate locations being considered due to the computational cost consideration.

Thanks for considering Subread and good luck for your project.

Wei

        

      
      
    

  
  


  


      

    

    
1 day later

    

      
      

        


  

        

      Bhagwat, Aditya
    

    

      
      #
    

  


  

      
Thank you Wei,

I actually love Rsubread, and use it with much appreciation on RNAseq projects, thank you for its creation :-)

Cheers,

Aditya