This works in updates to wfmash and default parameter settings that improve performance on variant calling tests.
Significant changes:
default to -s 10k-p 90 which is provides better alignment contiguity
update wfmash using biWFA for low-memory reduction-free (exact) patching of the alignment traceback, "erosion" of the traceback prior to patching has been enhanced, resulting in slower but more accurate alignment (as measured by variant calling tests from graph)
default to -k 19 which prevents underalignment and unfolded bubbles (and improves gfa2evaluation f-scores)
update smoothxg to pad a minimum of 311bp on each flank of POA problems but set a low default POA overlap -O 0.001 which basically sets the overlap length to a fixed 311bp
iterate smoothxg 3x using much smaller POA lengths than previous---the idea here is to slowly condense the -k 19-sparsified seqwish graph, progressively closing large redundant bubbles. In testing, longer -G settings do not appear to help, and it appears to be best to use a shorter setting when applying local alignment in POA.
This works in updates to wfmash and default parameter settings that improve performance on variant calling tests.
Significant changes:
-s 10k
-p 90
which is provides better alignment contiguitywfmash
using biWFA for low-memory reduction-free (exact) patching of the alignment traceback, "erosion" of the traceback prior to patching has been enhanced, resulting in slower but more accurate alignment (as measured by variant calling tests from graph)-k 19
which prevents underalignment and unfolded bubbles (and improves gfa2evaluation f-scores)smoothxg
to pad a minimum of 311bp on each flank of POA problems but set a low default POA overlap-O 0.001
which basically sets the overlap length to a fixed 311bp-k 19
-sparsified seqwish graph, progressively closing large redundant bubbles. In testing, longer-G
settings do not appear to help, and it appears to be best to use a shorter setting when applying local alignment in POA.