How To Use The Recon Package To Identify Repeat Families From Genomic Sequences

Question

[Nathan Fig]

I'm trying to learn RECON and am experimenting using chr22. My steps so far, roughly:

1. Make blast database from chr22.fa
2. BLAST chr22.fa against its own database
3. Run MSPCollect.pl (RECON provided script) to create an MSP file

4. Run recon.pl on the MSP file and a list of sequence IDs

   However, blasting a sequence against its own database takes a prohibitively long time or results in 100% self hits. If I remove the self hits, I'm left with a bunch of alignments that RECON is then happy to work with, but I had to write my own Python script to filter those out.

   By now this is all feeling very convoluted, so my question is: Am I even close to doing the BLAST part correctly? The RECON home page says to avoid self hits for performance sake, but I haven't been able to discover how. Any other glaring mistakes?

[JC]

maybe you need to use something more complete like RepeatModeler (http://repeatmasker.org/RepeatModeler.html) or Repet (http://urgi.versailles.inra.fr/Tools/REPET).

[Istvan Albert]

I have sent an email to the maintainer of this package via the request help link above.

[Nathan Fig]

Ah neat feature; thanks.

Answer 1

[SES]

You want to avoid self hits because it complicates the graph structure and doesn't add any useful information. As you know, comparing a sequence to itself will result in all self hits, and it will take a very long time for sequences that are megabases in length. My advice would be to split the input into overlapping sequences and run blast on subsets of that in parallel. If you output your blast as tab-delimited it will also save space. Since the blast to MSP conversion script provided with RECON takes only a standard blast report I am providing a script below to convert a tab-delimited blast to the RECON MSP format.

This assumes you have many input sequences, so comment line 16 as indicated above if doing a self comparison of one sequence. As JC mentioned in the comments, RECON is likely not the best program for your task (it is buggy, seg faults for silly reasons, produces a very high percentage of artifacts, etc.), but I hope this helps someone.