Assembly contiguity & sequencing depth

[TymekPieszko]

Hello,

I have two genomic libraries with mean read depth of 24x and 120x. Surprisingly, after megahit assembly, the lower-depth assembly is more contiguous (twice as few contigs, twice as high N50). Are there any possible explanations for that in terms of the assembly algorithm? Subsampling reads from the high-depth library again gives improved contiguity - what's going on?

All the best, thanks for any help in advance!