Polishing errors close to window borders

[fmlt]

Hi
We are evaluting the quality of raven assemblies using Nanopore V14 raw reads and we noticed errors close to window borders that we belive may be caused by racon. More specifically, after mapping the reads to the polished unitigs, we can find places where almost all reads suggest that the assembled/polished sequence should be different.

We used a basic variant detector to find where these errors occur and they seem to happen "close to" the window border (it is hard to have a clear conclusion because there is no direct way to know where the borders though). The following tables show high frequency variants with windows of 300, 500 and 700 bp length.

300

500

700

We tried a two-step polishing approach using 500 and 700 windows on the first and second step without better results. The "--no-trimming" option also did not help. Is there anything you can suggest to improve such errors?

Thank you in advance for the help.

[JWDebler]

Interesting, because I just noticed something when the --no-trimming option is used (#77) (due to removal of telomeres otherwise), that BUSCO scores drop significantly. Maybe the introduced errors you show lead to broken gene models and therefore lower BUSCO scores.