Metavelvet issues. meta-velvetg.Graph2-stats.txt missing? Assembly not reproducable?

Question

[JV]

Hi,

I'm using meta-velvetg version 1.2.02 for my Illumina metagenome assembly. I did multiple assemblies with different kmer values. The assemblies seem to have finished without any error messages, and I have "meta-velvetg.contigs.fa"-files as results. However, I want to follow the recommendations on the metavelvetg-website that suggest manually adjusting the khmer-coverage values after a first assembly run and then reassembling with meta-velvetg.

However the instructions state that meta-velvetg.Graph2-stats.txt should be used for this

My assemblies did not produce this file. All the result files I got are these:

contigs.fa
Graph2
LastGraph
Log
PreGraph
Roadmaps
Sequences
stats.txt
meta-velvetg.asm.afg
meta-velvetg.contigs.fa
meta_velvetg.contigs.fastainfo.txt
meta-velvetg.LastGraph
meta-velvetg.LastGraph-stats.txt
meta-velvetg.split-stats.txt
UnusedReads.fa

I wonder if this may indicate that something may be wrong with the assembly-run?

Also I got slightly different results when recalculating assemblies. The paramaters and input data were all identical in both assemblies, only the number of available openMP-CPU cores changed (8 cores in the first assembly, 10 in the second).

my results with 8 cores:

statistics for meta-velvetg.contigs.fa
    numcontigs : 8621998
    minlen : 83
    maxlen : 32993
    totallen : 1668277937
    avlen : 193.490874969
    medianlen : 154.5
    n50 : 176
    number of contigs larger or equal to:
        500 : 244632
        1000 : 54292
        2500 : 6693
        5000 : 1044
        10000 : 132

My results with 10 cores:

statistics for meta-velvetg.contigs.fa
    numcontigs : 8634001
    minlen : 83
    maxlen : 32993
    totallen : 1669579044
    avlen : 193.37257941
    medianlen : 155
    n50 : 176.0
    number of contigs larger or equal to:
        500 : 244260
        1000 : 54044
        2500 : 6690
        5000 : 1029
        10000 : 139

Are these differences normal, or should meta-velvetg produce exactly the same results when reassembling identical datasets?

Answer 1

[JV]

in case anybody wonders:

The problem was that I stated "-exp_cov auto" in the command for meta-velvetg. Simply leaving this argument away (only for meta-velvetg. You still have to give this argument for velvetg) solved this problem and even yielded significantly better assemblies even before coverage optimization.