misassembly introduced in hifiasm v0.5

[chhylp123]

misassembly introduced in hifiasm v0.5

[chhylp123]

Thanks so much for point that. It seems that the mis-assemblies are caused by the our new purge_dup. I will expose its parameters to users today, and let you know when it is available.

[lh3]

v0.5 fixes one misassembly on our data, but generally hifiasm does produce large misassemblies occasionally. @chhylp123 will expose some purge_dups parameters, which may help. We are also thinking about the possibility to integrate some bogart heuristics in future.

[chhylp123]

I have exposed three purge_dups parameters: '-l', '-s', '-O'. I guess it may help for fixing assemblies. Please use the latest commit with version 0.5-dirty-r247 (hash: 7f6725e). Looking forward to your results @HenrivdGeest .

[HenrivdGeest]

Thanks, I will start a sweep asap. Is their a way to re-use the error corrected reads? With some parameter testing, it can be faster to run that step once.

[lh3]

If you specify the same "-o prefix" option, hifiasm will reuse "prefix.*.bin" files and skip error correction and overlapping. Note that older assembly graphs will be overwritten if you do this.

By the way, what is the heterozygosity of your genome? v0.5 outputs multiple k-mer histograms in stderr. Is it possible to show us the first histogram? Thanks.

[HenrivdGeest]

I am running the loop now. The first kmer plot:

I also ran genomescope on kmer=21 on the hifiasm error corrected reads ( on more data, so the coverage peaks are at different numbers) before:

Here you see clearly 4 peaks. The hifiasm just 2, or barely 3, but that kmer is at 51, meaning that 2 haplotypes might already get completely separated at that size?

[lh3]

Thanks a lot! Is this public data? We have been mostly using animal data for testing. The heterozygosity is much lower in comparison to some plant genomes. While redwood heterozygosity is high (higher than your genome) and its data is public, it takes several days to assemble, making testing very difficult. If your data is public or can be shared with us privately, it would be ideal for development purposes. Some hifiasm parameters are tuned for low heterozygosity, we may have a lot of room for improvement.

Also, what is the preferred output for your genome? Is it one primary assembly representing one haplotype?

[HenrivdGeest]

I made 18 assemblies, all combinations of -l (0,1,2) -s (0,75,0,90) and -O (1,5,10). But all of them show exactly 2 misassemblies. I measured my version 0.3 the same, and that one has none. The n50's of 0.3 and the "v0.5 -l0,-s0.75 -l1" version are almost equal:

@lh3; My haploid genome size is around 400-600Mb, I was never able to collapse my assemblies to lower than 800Mb with purge_haplotigs, so I think some haplotypes are very heterozygous, and some not so, and can be easely collapsed. (hence the reason why I also never get a 2Gb genome fasta). Our plant is called a 'segmental allopolyploid' (so the other parts are autopolyploid).
Regarding what I want as output, in this case I want to have the unitigs, as much separating the haplotypes as possible. (I know all my previous remarks are about the contigs). The unitigs are upto 1.6Gb in size, but still contain collapsed haplotypes, probably because there are no snps to be found between alleles. This data is confidential, and can therefore, unfortenately, not be shared.

[chhylp123]

I see. Thanks a lot. Let me check what's the difference between v0.3 and v0.5...
BTW, could you please run v0.5 on the bin files generated by v0.3?

[chhylp123]

I think you can have try to further increase -s and -O. I guess even '-s 0.99' is fine for hifiasm. Actually '-s 1' doesn't mean hifiasm only purges exactly same haplotigs, it still allows differences.

[HenrivdGeest]

I tried increasing -s to 0.999 and -O to 75, but that did not made any difference in the output regarding the misassembly. I tried running v05 on the v03 bin files, but for v03 I only have a *reverse.bin, and that alone doesn't seem to work, I think it starts assembling from scratch.

[chhylp123]

Thanks a lot, I will expose another option to users. I believe that can avoid these two mis-assemblies.

[HenrivdGeest]

BY any change any luck on these new options?

[chhylp123]

Oh, I'm so sorry I forget that : ( . I will expose it this day.

[chhylp123]

Please wait me one more day, I will fix it soon. Thanks a lot : )

[chhylp123]

I have exposed '-u' to disable post-joining (0.7-dirty-r256). Hope it is helpful. I'm so sorry for the deay : (.