SPAdes 3.7.1 recommendation

[sjackman]

WARNING: SPAdes version 3.9.0 is being used. It will work, but better results are usually obtained from Circlator using SPAdes version 3.7.1. Although 3.7.1 is not the latest version, we recommend it for Circlator.

Do you have any thoughts as to why SPAdes 3.7.1 performs better than 3.9.0 with Circlator?
cc Anton Korobeynikov @asl

@ilovezfs and I maintain the circlator package for Homebrew/Science, and we'd prefer to use SPAdes 3.9.0 with Circlator if possible.

[martinghunt]

If it's a pain, go with 3.9.0.

I tested all versions from 3.6.2-3.9.0 on the 14 NCTC samples from the Circlator paper. SPAdes 3.7.x resulted in the most circularizations. The differences were mostly in the samples with several plasmids. But the differences were in the SPAdes reassembly of contig ends, not SPAdes itself assembling the plasmids and calling them as circular. Don't know any more than that. I just looked at the final counts of number of circularizations, haven't dug into the details of the reassemblies and what was different about them. Unlikely I'll get the time to.

That said, I wouldn't be surprised if a different dataset gave different results anyway, as assembly can be so variable. So I'm ok with 3.9.0 in homebrew. The same thing will happen with debian packaging anyway.

[sjackman]

Great. Thanks for your quick response, Martin.

[asl]

The results should be pretty much the same. The major differences were in metaSPAdes, plasmidSPAdes and rnaSPAdes pipelines and almost nothing was changed for normal isolate assemblies besides several bug fixes in the repeat resolution. So, it may happen that SPAdes 3.7 indeed did unsafe decisions (probably even introducing misassemblies), however, providing longer contigs.

[martinghunt]

Interesting. Thanks for the info @asl. It's probably the bug fixes in repeat resolution that made the difference. Circlator is only reassembling contig ends and complete small contigs - all of which are likely to contain repeats.

[asl]

Reassembling just the end of the contigs does not make much sense since SPAdes tries to remove overlaps from the repeats and report the ends of repetitive contigs only once. The proper (and accurate way) is to operate on .paths / .fastg files

[splaisan]

where can we get the old version of spades, I do not find it on their site. Could you possibly post it on your git as archive?
Thanks