Amplicon assembly

[BBthekid007]

I'm trying to sequence viral glycoprotein genes using a Moleculo-like method involving clonal amplification, followed by tagmentation with NexteraXT and sequencing on the MiSeq. Each of the clonally amplified genes will be uniquely indexed, so I'll be able to demultiplex all of the short reads that correspond to a single input transcript. The sample prep is pretty straightforward, but I'm not sure how to assemble the resulting short reads.

The glycoproteins are pretty variable, so it might be difficult to use a reference for the assembly -- since the gene is only about 3kb, would it be better to use de novo assembly? If so, would something like Velvet or SOAP work, or are there better assembly programs designed to assemble gene-sized fragments?

-Bryan

[swbarnes2]

I'm trying to sequence viral glycoprotein genes using a Moleculo-like method involving clonal amplification, followed by tagmentation with NexteraXT and sequencing on the MiSeq. Each of the clonally amplified genes will be uniquely indexed, so I'll be able to demultiplex all of the short reads that correspond to a single input transcript. The sample prep is pretty straightforward, but I'm not sure how to assemble the resulting short reads.

The glycoproteins are pretty variable, so it might be difficult to use a reference for the assembly -- since the gene is only about 3kb, would it be better to use de novo assembly? If so, would something like Velvet or SOAP work, or are there better assembly programs designed to assemble gene-sized fragments?

-Bryan

I think velvet will work, but you may want to downsample; velvet might not like it if your coverage is huge, which could be the case with such a tiny sample.

Rather than wrestle with velvet on all the samples, maybe you could align first, and if the alignment looks bad, then velvet.

With such a small genome, you could align, fix SNPs, and align again a few times, and that wouldn't take very long.

[BBthekid007]

In case anyone is interested in how this turned out, I mocked up a dataset of error-prone pseudo-reads using a known viral gene sequence. The reads were 250bp in length, to mimic single-end MiSeq reads. Using 100k reads, which would correspond to coverage of about 9000X, de novo assembly with Velvet worked great -- produced a single contig that identically matched the original gene.