Cleaning up trinity assemblies

Question

[dpearton]

Hi,

I have performed de novo assemblies on mRNA using trinity with PE illumina data (125bp PE). Trinity gives a large number of contigs (around 250,000) but many of these are very short (200-500bp). In addition there can be many versions of the same contig - either different isoforms, splice variants or variant assemblies. Is there a way to "rationalise" or collapse the assembly? For example only taking the longest isoform of each contig? I know that would possibly be throwing away any info on splice variants but that is not something that I'm too interested in at the moment. And/or having a size cut-off?

Assembly stats are as follows:

################################
## Counts of transcripts, etc.
################################
Total trinity 'genes':  185527
Total trinity transcripts:      252342
Percent GC: 40.60

########################################
Stats based on ALL transcript contigs:
########################################

        Contig N10: 4739
        Contig N20: 3372
        Contig N30: 2579
        Contig N40: 1972
        Contig N50: 1452

        Median contig length: 406
 Average contig: 803.80
        Total assembled bases: 202831907


#####################################################
## Stats based on ONLY LONGEST ISOFORM per 'GENE':
#####################################################

        Contig N10: 4338
        Contig N20: 2823
        Contig N30: 1957
        Contig N40: 1258
        Contig N50: 802

        Median contig length: 357
        Average contig: 621.99
        Total assembled bases: 115396193

Thanks,
Dave

Answer 1

[seta]

If the short contig is not your interest, you can easily apply the flag of --min_contig_length 400 or 500, for example. However, be careful about it as some of protein sequences have short length, then you may miss them. Although there is a script to get the longest isoform, the longest transcript is not always the best one, so you can consider filter the lowly supported transcript using RSEM output. Hope this helps.

[dpearton]

Hi,

I know the longest might not be the "best" but I'm not sure what other criteria to use. I imagine that reads/contig normalised for length might be useful but I have no idea how to implement that. I've visualised my assemblies in tablet and there is a wide range of reads/contig.

I've not used RSEM - how would this work without a reference genome?

[seta]

I usually use the min_contig_length 300 to get rid of many short contigs, you can type just --min_contig_length 300 along with your trinity command. About RSEM, please use the align_and_estimate_aboundance.pl script within Trinity package then using RSEM output, you can filter contigs with fpkm less than 1. You can take a look at http://trinityrnaseq.sourceforge.net/analysis/abundance_estimation.html