Genes with multiple copies assembling as single contig

[cyanoevo]

Hi all,

I'm doing a denovo assembly of a cyanobacterial genome with SPades, all is working well but when there are multiple copies of a gene (e.g. 16srRNA gene), it appears that all reads associated with that gene are being mapped to a single contig.

Coverage of these contigs appears to correspond quite well to number of expected copies in the genome (i.e. normal coverage ~50x, for a contig with a gene with four copies, coverage ~200x).

Does anyone know of a method to prevent this from happening so that each of the copies assemble separately in different contigs?

Cheers
N

[ecastron]

Hi Cyanoevo,

I have the exact same problem. Did you ever find an answer?

Cheers,

Eduardo

[Brian Bushnell]

You can try mapping reads to a 16S copy, then clustering the reads that mapped, then assembling the clusters. This will work if the reads are sufficiently long (for Illumina, merging them may be useful) and the 16S are sufficiently different. If not, you'll just get one cluster. You probably need overlapping 2x250bp reads at a minimum (insert size around 400bp+) to have a good chance.

You can cluster like this with Dedupe (packaged with BBMap):

dedupe.sh in=merged.fq -Xmx30g am=f ac=f fo c rnc=f mcs=50 mo=350 pto pattern=cluster_%.fq

The "mo=350" specifies a min overlap of 350bp. This should be around 80%-90% of your read length. If you have single-ended 250bp reads, set it to 200; if you have merged reads with an insert size of around 400bp, try 350. If you have 100bp non-overlapping reads, don't bother, they're too short.

For this kind of situation, which is very sensitive to chimeras, I recommend merging reads with BBMerge using the "vstrict" flag.

[ecastron]

Thanks Brian, I'll give it a try. I anticipate that I'm going to get one cluster because the reads are seemingly identical. It's suggestive that the coverage for the rRNA operon is about 3 times the coverage of the neighboring genes so at a minimum I'll report that in the submission.
I guess the alternative would be going back to the wet lab to check how many copies there are.

Cheers,

Eduardo

[akorobeynikov]

Thanks Brian, I'll give it a try. I anticipate that I'm going to get one cluster because the reads are seemingly identical. It's suggestive that the coverage for the rRNA operon is about 3 times the coverage of the neighboring genes so at a minimum I'll report that in the submission.
I guess the alternative would be going back to the wet lab to check how many copies there are.
Eduardo

Since you're having multiple copies of 16s then you need to have sufficiently long insert length to allow assembler resolve this repetitive region. Otherwise, indeed, everything will be inside single contig. Given the length of 16s you'd need at least mate pairs of > 2-3kb insert length or long reads (PacBio / Nanopore).

[ecastron]

Thanks for the reply! That was my impression; that I wouldn't be able to resolve it with 300bp insert library but only with mate pairs or long read technology.

Cheers,

@ecastron