assembly with unsatisfying results - use new reads with larger inserts?

[martin_313]

We recently finished an assembly of Guignardia bidwellii paired end DNA reads using Abyss and refined the results with SSpace. However, we were not quite satisified with the results (still to many contigs).
We are now considering generating new libraries with larger insert sizes between paired ends to resolve repeats. Is there a genereal guideline for choosing those insert sizes? From what I understood so far, an optimal insert size would be larger than the length of the largest repeat appearing in the genome (is this correct?).
Sadly, we do not have too much information about the genome for estimating the size of repeats in the genome. Is there some way to make an eductaed guess based on the Abyss-output files (like the contig linkage graph) of our last assembly? Or should we simply go for the largest insert size possible?

Thank your for answers
Martin

[peromhc]

before talking about making new libs, tell us a little about the current libs and sequencing

How many libs do you currently have, what insert sizes?
How deep is your sequencing currently--10X, 20X, more?
How big is your genome.

[MadsAlbertsen]

The first thing I do after an assembly is always a contig length vs. coverage plot to estimate the repeat content. It's usually quite easy to see the repeats.

rgds
Mads

[martin_313]

Thank you for your answers.

peromhc, since we received the libraries in a single file without too much information, i don't know the exact number of libraries. The insert sizes range from 100 to 500 with the majority between 200 and 400.
All reads were 101 bp long.
The average sequencing depth seems to be around 50. The size of the genome is estimated to be around 40 mbp. Our assembly contains around 30 mbp.

MadsAlbertsen, how do you get such a plot? What program/command do you use?

[MadsAlbertsen]

Most assemblers have a function to obtain the average coverage for each contig - it's a long time since I've used ABySS so I can't remember if it has an output contig coverage option.

Otherwise map the reads back to your contigs using any short read alligner and calculate the coverage.

You can easy do the coverage vs. contig length in excel. But if you have loads of contigs the smoothScatter function in R is great.

rgds
Mads