little help with abyss assembler

Pavle Eric

unread,

Aug 9, 2023, 11:22:56 AM8/9/23

to ABySS

Hello everyone,

Recently I got some sequencing data, pair-end 150bp Illumina reads of a bird species that has not been assembled to this date. So I tried your assembler with a few k values (K=69, 79, 89, 99, 109, 119). Out of these values, K=99 gave the best assembly parameters, and a busco score of 94.4%. Also, the genome size is very close to the genome size of a species from the same genus, that has chromosome level assembly on NCBI. On the other hand, I am aware that when using only short reads, the genomes are going to have a lot of gaps, but having over 500k scaffolds seems a lot.

Wanting to try something else, to make a better assembly, I read in one of the discussions here, that you recommend pair merging before the assembly so I wanted to try and get an even better assembly.
First, I tried your abyss-mergepairs for overlapping pairs, but since it is single threaded it was taking too long, so I tried pear.
My understanding from the discussion I read, is that I should first merge overlapping read pairs with (abyss-mergepairs or pear), and then feed the unassembled read pairs
to konnector so it can merge non-overlaping read pairs. (after Konnector only a few hundred megabytes of data remained (out of 150GB per F and R reads).
My next understanding is that I should feed the abyss-pe with both the raw reads as pair-end reads as well as the two sets of assembled reads and remaining unassembled
reads from konnector as single-end reads. After doing so my assemblies were significantly worse than assemblies without pair merging as they had more than double the number of scaffolds (I did this with k values 99 and 95).
I have assembly statistics included here as well as the commands I used in the process.
Am I doing something wrong in the merging process or is it just possible that the basic assembly is giving better results?

n n:500 L50 min N75 N50 N25 E-size max sum name
1172362 43763 5419 500 34143 64748 111937 81861 503470 1.19e9 ACH01_abyss_k95-scaffolds.fa *with merged pairs
1114044 43155 5280 500 34787 66368 114968 84482 700626 1.189e9 ACH01_abyss_k99-2-scaffolds.fa *with merged pairs
541770 92136 13320 500 13550 25915 45024 33741 242779 1.185e9 ACH01_abyss_k119_3-contigs.fa
510987 29948 3211 500 55726 107529 191111 138094 751714 1.188e9 ACH01_abyss_109-scaffolds.fa
610357 29148 3167 500 56070 109390 192459 140203 981356 1.188e9 ACH01_abyss_99-scaffolds.fa
733825 29529 3281 500 53768 104471 185035 135435 958488 1.19e9 ACH01_abyss_89-scaffolds.fa
916489 30935 3600 500 48840 95807 168852 123747 806307 1.192e9 ACH01_abyss_79-scaffolds.fa
1169123 35922 4568 500 39426 76236 133473 98583 780728 1.196e9 ACH01_abyss_69-scaffolds.fa

pear -f ACH01_1.fq.gz -r ACH01_2.fq.gz -o ACH01 -j 63

konnector -j 63 -k 95 -b30G -E -D8G -v --fastq -o k95 ACH01.unassembled.forward.fastq ACH01.unassembled.reverse.fastq

abyss-pe B=320G k=109 j=63 l=40 s=1000 v=-vv name=ACH01_abyss_k109> ACH01_abyss_k109.log 2>&1 lib='pe1' pe1='../ACH01_1.fq.gz ../ACH01_2.fq.gz' se='../ACH01.assembled.fastq.gz ../k95_pseudoreads.fq.gz ../k95_reads_1.fq.gz ../k95_reads_2.fq.gz'

Best Regards,

Pavle Eric

Lauren Coombe

unread,

Aug 9, 2023, 11:34:31 AM8/9/23

to ABySS

Hi Pavle Eric,

A couple follow-up questions so that I can best help you.

- What is the difference between your read files "../ACH01_1.fq.gz ../ACH01_2.fq.gz" and "../k95_reads_1.fq.gz ../k95_reads_2.fq.gz"? Just trying to understand why one set is specified as paired-end and the other is not.

- For a given k-value, could you provide the stats for unitigs, contigs and scaffolds for both the original and merged pairs versions? That will help me to see the improvement at each step (since the PE information is used post-unitig stage).

In general, 94.4% BUSCOs is a really great result! Are there any plans to do any further sequencing with a different data type (ex. linked reads, long reads, etc)? I just ask because further scaffolding with longer-range evidence could be beneficial depending on how repetitive this genome is.

Thanks for your interest in ABySS!
Lauren

Pavle Eric

unread,

Aug 9, 2023, 12:59:46 PM8/9/23

to ABySS

Hello Lauren,

Firstly thank you for a very quick reply. So ACH01_1.fq.gz and ACH01_2.fq.gz are F and R raw reads and k95_reads_1.fq.gz, k95_reads_2.fq.gz are leftover unmerged reads that konnector could not merge (they are a few hundred megabytes large, before the gunzipping, (although they are not the same size which was a little odd)). I used k=95 for konnector pair merging. I am sending you assembly statistics for untigs, contigs, scaffolds stage of each assembly.
As for further sequencing with longer reads, I would very much love to have that data, but I think that it won't happen in the near future. As for how repetitive the genome is, how informative can a closely related species genome be helpful with that? The data I am working on is Aquila heliaca and there is a chromosome level assembly of Aquila chrysaetos. Could its genome be helpful with scaffolding or resolving some uncertain regions?

abyss-fac ACH01_abyss_69-unitigs.fa ACH01_abyss_69-contigs.fa ACH01_abyss_69-scaffolds.fa |tee ACH01_abyss_69-stats.tab