Inconsistency between file out_corrected.fa and out_srfilter_CSE.out

[Zhenlisme]

Hi,

I used the CRAQ to evaluate my hifiasm assembly. My command line is

perl ~/CRAQ/bin/craq --genome scaffolds.fa --sms_input minon_all.trim.fastq.gz --ngs_input total.1.fq.gz,total.2.fq.gz --report_SNV T --break T --map map-ont --thread 30 -D scaffold_craq

I found an inconsistency between the file out_corrected.fa and out_srfilter_CSE.out. There are 4546 lines in out_srfilter_CSE.out file, representing 4546 CSE points that need to be broken in the draft assembly. However, there are only 2099 scaffolds in the file out_corrected.fa, meaning CRAQ did not break all CSE point recorded in the out_srfilter_CSE.out file.

I am wondering if this is normal and what criterials CRAQ is using to break misjoint scaffolds.

grep “>” out_corrected.fa|head

head out_srfilter_CSE.out

Thanks

[JiaoLaboratory]

Thank you for using CRAQ. Yes, this is reasonable. If several error sites are very close to each other, such as clustered within a 10k window, these errors likely originate from a structural error region. CRAQ will choose to break the contig at the most probable error location. Can refer to the out_final.CSE.bed file for details.

[Zhenlisme]

Thank you for your reply.
Now I understand that CRAQ clusters break sites together by distance. In my case, the window size CRAQ used for clustering is likely much longer than 10 kb, probably longer than 600 kb. Would that overestimate the genome assembly quality? Because 600 kb is a bit long, which can comprise many misjoins. Does CRAQ provide any parameters to restrict the window size to be used in the clustering process? Could you please give me any suggestions?

Best,
Zhen

[JiaoLaboratory]

Thank you. Unfortunately, CRAQ currently does not include a user-defined parameter for clustering error hotspots, which requires users to manually process the out_srfilter_CSE.out file according to their needs. Additionally, empirically, if the distance between two CSEs is too large (e.g., >100 kb, exceeding the length of a typical ONT sequencing read), forcibly applying the clustering step may not be reasonable, as it could overestimate genome quality.

Best,