Assembly quality using different k-mer sizes

[ShailNair]

Hi,
I assembled pair-end metagenomic reads via spades. I used two sets of k-mers to evaluate the assembly quality. The raw reads were 250bp long.

Assembly-1 with k-mers- 21,33,55,77
Aseembly-2 with k-mers- 21,33,55,77,99,127

Here is the Metaquast statistics:

Statistics without reference	Assembly-1
contigs	38417
contigs (>= 0 bp)	288336
contigs (>= 1000 bp)	14500
contigs (>= 5000 bp)	2978
contigs (>= 10000 bp)	1471
contigs (>= 25000 bp)	583
contigs (>= 50000 bp)	313
Largest contig	1048960
Total length	119714142
N50	21147
N75	2648
L50	682
L75	5284
GC (%)	58.3
Mismatches
N's	0

Statistics without reference	Assembly-2
contigs	41525
contigs (>= 0 bp)	132369
contigs (>= 1000 bp)	16003
contigs (>= 5000 bp)	3174
contigs (>= 10000 bp)	1523
contigs (>= 25000 bp)	581
contigs (>= 50000 bp)	281
Largest contig	2737835
Total length	123551515
N50	16944
N75	2317
L50	875
L75	6492
GC (%)	58.34
Mismatches
N's	0

My confusion is, from these statistics which is the better-looking assembly?. Though from N-50 Assembly-1 looks better, it has less number of contigs than assembly 2. Also, the largest contig in Assembly-1 is smaller than that of Assembly-2. Or do I need to find the orf's, annotate, and then compare the results? (which will take some time as I have many reads)

[asl]

Neither N50, nor # of contigs or the length of largest contig is the "quality of assembly" (especially for metagenomes). It's very easy to obtainer longer contigs that contain many problems.

So you'd need to analyze your data further on to decide.

[ShailNair]

I see. I will analyze further and see how the differences are reflected in the two assemblies.

Thanks ☺️