[Corydoras]

Thank you!!

[Corydoras]

Dear RAD/Assembly community,

I was hoping I could pick your brains again regarding my assembly.
I normalized coverage and together with using velvetoptimiser I am quite happy with the results I am getting! For instance, I assembled the reverse reads only, which gave me ~33,000 contigs (this is roughly what I am expecting based on preliminary data) with an N50 of 400. Kmer coverage distribution looks good, too.
However, I now wanted to feed these in as 'long reads' alongside my raw data in velvet to resolve repeats better and connect the forward and reverse reads properly. Weirdly, my N50 goes down to about 380? I have visualized the assembly in tablet, and also mapped back some of my raw data (using BWA) back to check whether forward and reverse reads were actually assembled correctly, and for the few contigs I surveyed by eye this seems to be the case.
I am not sure whether I need to be worried about the decreasing N50, this really seems to be a bit counter-intuitive, as my contigs surely should increase in size by quite a bit when assembling forward and reverse as opposed to just the reverse?
I was also wondering whether there is a good way of quality filtering my contigs from my final assembly. For instance, I only want to keep contigs for future analysis to which both forward and reverse reads map to.

Many thanks in advance for any pointers!!