Contig count explodes from hundreds to 4000 after LongStitch pipeline

[rubadevis]

I'm running the complete LongStitch pipeline on a mammalian genome (~2.5Gb) using standard parameters. However, my final contig count jumped dramatically from a few hundred (pre-LongStitch) to 4000 contigs, despite maintaining excellent assembly metrics:

N50: 73Mb (chromosome-scale)

Issues:
Is this expected behavior for mammalian genomes with moderate repeats (~40-48% LINEs)?

How can I reduce contig count while preserving N50 73Mb?

Parameter tweaks for LongStitch to favor scaffolding over fragmentation:

[lcoombe]

Hi @rubadevis,

If you look at the length breakdown of the sequences, I'd guess that many of the sequences are very short, even 1 base long. That is because of the Tigmint misassembly correction step - it is very common to have very short sequences generated when the tool cuts the assembly at misassemblies. I explain more in this issue: #17

If you haven't already, I would first recommend looking at the the number of contigs with a lower length threshold (ex. 1kb) to see if you observe a different trend.

Thank you for your interest in LongStitch!
Lauren

[rubadevis]

Yes, I got you. It is due to the misassembly correction performed by Tigmint. I ran the tool using standard parameters, with both span and dist set to auto.

My goal was to use Tigmint for misassembly correction while also reducing the number of contigs in the assembly. However, I observed that the contig count did not decrease as expected.

Could you please advise:

Is there a recommended way to reduce the number of contigs when using Tigmint?
Would it be appropriate to remove very short contigs/reads after running Tigmint to help reduce overall contig counts for downstream analysis?

Any guidance on best practices for improving assembly contiguity in this context would be helpful.

[lcoombe]

Hi @rubadevis,

I think it's important to distinguish contiguity and the number of contigs in the assembly. Usually, you would use the N50/NG50 of the assembly to assess contiguity as you mentioned in your initial comment, as opposed to simply the number of contigs. You can have a very contiguous, chromosome-level assembly with small sequences that contribute very little to your assembly reconstruction - it is common practice to filter those sequences out anyway.

Because Tigmint is a misassembly correction utility, by design it will always increase the number of contigs - since it is breaking the assembly at misassemblies. The next step(s) of the LongStitch pipeline after Tigmint is scaffolding using the same long reads, which then puts the contig pieces back together.

So yes, it is common practice to remove small sequences (for example, < 1kbp or similar) from your final assembly. You can track the NG50, number of sequences and sum before and after this filtering - the NG50 and sum should be minimally impacted by the filtering.