1
$\begingroup$

I have developed a software for de novo genome assembly. Its performance varies gradually according to how much data you employ. At initial stages it often produces contigs that look like that when aligned to a reference genome:

s-aligner assembly of SRR8357422 Initial-stage s-aligner assembly of SRR8357422 Genome fraction:98,8% - Largest Alignment: 61.544 - NGA50: 59.409 - LGA50: 2

If I add more data, the contigs end up being larger, but that obviously takes more time. My question is, how does this result compare to, for example, one like that?

SPAdes assembly of SRR8357422 SPAdes assembly of SRR8357422 Genome fraction:93,46% - Largest Alignment: 128.391 - NGA50: 128.391 - LGA50: 1

Note how you can construct a contig covering almost the full reference genome assembling green contigs in the first case. Therefore, a blast search of genes will likely let you find more genes given that it covers more genome fraction.

Apart from determining what option is more convenient, I would also like to know how much more convenient is to have both results. Or in which cases it's more convenient.

Last (but not least!), is there a way to generate automatically larger contigs from these overlapping contigs in assembly 1? I have tried using miniasm with no success. Even minimap2 seems not to find overlaps well enough. I am developing my own algorithm also for that but would like to know if there is another way.

[EDITED on 17/12 to add another example with real data for both assemblies]

Improve this question
$\endgroup$
7
  • $\begingroup$ Hi, could you please clarify your question a bit more; it's unclear to me. Asking "which one is a more convenient assembly?" suggests to me that there should be at least two comparisons, but you've only provided a figure for one. There is also additional information needed about the assembly and reference genome (i.e. are you doing kmer-based assembly; is the genome highly-repetitive; does it have Ns in the sequence). $\endgroup$
    – gringer ♦
    Dec 16, 2020 at 23:01
  • $\begingroup$ Hi @gringer, it's not a question for a single case. It's something that happens in different assemblies in different circumstances. It's a generic question. The example in the image is real and corresponds to a run in PRJEB32127 but I have not assembled it with another software yet. I mean, in case I obtained an assemblie with a single contig but lower coverage (for example a contig 100k long) , which one of the two assemblies would be more useful. $\endgroup$
    – juanjo75es
    Dec 16, 2020 at 23:20
  • $\begingroup$ Maybe it's more clear with a more extreme case. Suppose it's a bacterial assembly (~2Mbp) and SPAdes get a contig with 3Mbp bit just covering 90% of the reference (the rest is chimeric) while the otrher assembler gets lots of contigs around 200k bp long but covering 99%. I'd like to know how important is to get larger contigs or if coverage and percentage of genes covered is more important. $\endgroup$
    – juanjo75es
    Dec 16, 2020 at 23:24
  • 1
    $\begingroup$ I will lsee if I find a real case with real data... $\endgroup$
    – juanjo75es
    Dec 16, 2020 at 23:25
  • $\begingroup$ I guess it's "genome fraction" where I said "coverage"... $\endgroup$
    – juanjo75es
    Dec 16, 2020 at 23:43

1 Answer 1

Reset to default
0
$\begingroup$

I asked that on Reddit. I didn't get an answer to all my questions but I got an answer for how to assemble the overlapping contigs.

If I use Flye I can assemble these and get larger contigs, better NG50, and higher genome fraction.

enter image description here s-aligner + Flye assembly of SRR8357422 Genome fraction:95,95% - Largest Alignment: 134.103 - NGA50: 134.103 - LGA50: 1

Given that the data in the first assembly apart from being more useful itself in my opinion, is objectively more suitable to generate a better result in all senses with a simple additional step that takes a few seconds, my only possible conclusion is that it is overall a better result and therefore this sort of result will often be preferable.

Of course, there are other aspects to consider, like error rate, that are not considered here as it wasn't part of the question (mainly because I think Quast is not adequate for calculating it). But BTW, mismatches per 100kbp are also lower.

[EDIT] I want to add that this is not granted that in all cases Flye will generate a clearly better assembly.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.