Is there any alternative for CD_Hit to remove redundancy from asemmbled trinity output file?

Question

[rahmati.razieh83]

Hi everyone

I have a problem with reduction of redundancy from trinity output file. I have got an assembled fasta file from trinity containing 302000 contigs showing so much redundancy. I used CD_hit to remove redundancies and get unigenes. After using CD_Hit the number of contigs reduced to 240000 contigs showing lots of redundancies again. CD_Hit was not effective to achieve unigenes. Please give me advise how can I get unigens and remove redundancies?

Thanks

[Sej Modha]

Did you tweak the identity threshold using -c on the cd-hit?

[lakhujanivijay]

When you say

240000 contigs showing lots of redundancies again

How do you verify that?

Try using TGICL

Answer 1

[h.mon]

Trinity has a somewhat new script to construct "SuperTranscripts" based on the gene-to-isoform relationships and the sequence graph structure leveraged by Trinity during assembly. I think this will result in a better representation of unigenes than using cdhit.

$TRINITY_HOME/Analysis/SuperTranscripts/Trinity_gene_splice_modeler.py \
   --trinity_fasta Trinity.fasta

Answer 2

[Jake Warner]

Getting 'unigenes' from Trinity assemblies is tricky business. I've found that Corset performs better than CD-Hit. Another idea is to BLAST all the transcripts and group them by reciprocal best blast hit.

[h.mon]

LACE and Corset are tools from the same group. Initially I thought LACE would be the preferred tool, as it was developed more recently, but I was wrong: according to one of the authors of both tools, they should be equivalent for the purpose of doing gene-level differential expression analysis. As the Trinity Trinity_gene_splice_modeler.py is based on the same algorithm as LACE, it should also be equivalent to Corset.

Answer 3

[bcontreras]

We have used our own https://github.com/eead-csic-compbio/get_homologues successfully. In fact we benchmarked against CD-HIT on https://www.frontiersin.org/articles/10.3389/fpls.2017.00184/full

I believe the main problem is that isoforms with different exons or retained introns are not properly handled by CD-HIT, but can clustered correctly with GET_HOMOLOGUES-EST in most cases