Minimum contig coverage to keep

[sebl]

Dear all,

I am working on de novo assembly of bacterial genomes. Genomes were sequenced with Miseq 2x250 libs with an average coverage ~40X. Adapters were trimmed with cutadapt by the seq provider, then I performed quality trimming with BBduk and now de novo assembly with SPAdes 3.9.

I ran some genomes with and without the cov-cutoff <auto> option and I've been comparing both assemblies.

I've seen that for some genomes, the cov-cutoff option in auto mode still keeps some contigs (actually, I am looking at scaffolds) with low coverage (~1) whereas for other genomes the minimum is much higher (~14).

Genomes will be compared using various tools, including ANI calculation. I'd rather not introduce bias by throwing away too much data from some genomes and not from others, or the contrary (keeping "junk" in some and not in other genomes).

I BLASTed some of the low cov contigs to see if these are contaminations and I cannot tell if they are, because I they align to various sequences in the genus I am working on.

So the question is basically:

1. How SPAdes chooses the minimum coverage using the cov-cutoff <auto> option?

2. Is there and what is the minimum contig coverage people suggest to keep/discard contigs?

Thanks!

[Brian Bushnell]

Are these single-cell libraries or isolate libraries? With isolate libraries there is no reason to expect highly variable coverage and contigs with much lower coverage than normal are most likely contaminants (or "junk"). If they align to the same genus then they are less likely to be contamination and more likely to be ... well, some other variety of thing that you don't want, like assembly from chimeric reads or some other artifact. With single-cell it's much more difficult to determine.

[sebl]

These are isolate libraries.

When I BLAST short contigs (<200bp) with high coverage (10'sx), these are often parts of ribosomal genes. When I BLAST contigs of length 200-~500bp, where most of low coverage contigs are, these are often pieces of different species in the genus or so isolates not identified to species level.

So the question is then what is "coverage much lower than normal"? Should I stick to a standard cut-off for all genomes? What this cut-off usually is? Or the "auto" mode in SPAdes should be enough?

[Brian Bushnell]

"coverage much lower than normal" is on a per-library basis; you can't have a single cutoff value that's always appropriate. That said, while it's difficult to make strict rules... if a contig has coverage under 25% of the median coverage, in an isolate, it's probably best to discard it. Bear in mind, though, that bacterial isolates can have differential coverage by up to ~50% biased toward the origin of replication if you gathered the DNA during exponential growth phase.