When a chain is split during base-level alignment, its chaining score is
reduced. However, the chaining score of its suboptimal chain remains the same.
This leads to underestimated mapping quality.
In human/mouse, the GTr..yAG pattern occurs to 91/92% of all GT-AG introns.
Modeling r..y clearly leads to higher accuracy. However, in SIRV, this
percentage is reduced to ~60%. The default "--splice --splice-flank=yes"
leads to lower accuracy. If someone benchmark minimap2 on SIRV, this would be
bad, but minimap2 is developed for practical applications, not for benchmarks.
I will live with that.
This bug may lead to a wrong minimizer when a HPC k-mer is longer than 256bp.
When there is a seed match involving this wrong HPC k-mer, the correct seed
sequences do not match in fact. This violates the assumption in align.c and
subsequently causes a segfault, which is what #47 has caught. This bug lurked
in the earliest piece of code and affected all released minimap2 versions so
far. It is extremely rare and does not affect the prebuilt GRCh37/38 indices.
[PMID:18688272] shows that the base following GT tends to be A or G (i.e. R) in
both human and yeast, and that the base preceeding AG tends to be C or T (i.e.
Y). In the new model, we pay no cost to GTr..yAG, but we pay half of the cost
if there is no r or y. This improves the junction accuracy when mapping to
human and mouse and decreases the accuacy when mapping to SIRV. My guess is
that SIRV does not honor this trend. Need to investigate in future.
Also in this commit, --cost-non-gt-ag is aliased to -C. The default is changed
to 9 instead of 5. I also added --splice-flank to enable the above model. This
may become the default once I confirm my hypothesis on SIRV.
Heng Li
cwright
mvdbeek
martinghunt
Carlos de Lannoy