Explore alternate correction strategies
skoren opened this issue · 0 comments
skoren commented
Current correction works well but is quite slow due to the cost of overlapping, making up >50% of the total verkko runtime.
Options include:
- lja w/the --dimer-compress 1000000000,1000000000,1 option to avoid compressing dimers
- hifiasm
- faster potential overlap identification followed by edlib (aka the mhap strategy)
Results from testing so far on HG002:
- current correction - T2T ctg/scf 15/26, QV 53
- lja - T2T ctg/scf 4/10, QV 53
- hifiasm - T2T ctg/scf 8/15, QV 45
This would seem to indicate hifiasm is over-homogenizing the reads confusing the final assignment and thus consensus. LJA ran w/dimer compression so this likely hurt our ONT resolution alignment and needs to be re-run.