In-frame deletion being missed by moPepGen

[zhuchcn]

This case is caught by fuzz test with in-frame deletion. In the graph below, this in-frame deletion causing peptide RRA to be deleted. The node AP at the bottom is what this deletion produces, and has the same sequence as the AP on the top.

VMDASAFEIFSTFPPTLYQDDTLTLQAAGLVPK-AALLLR-AR-R-AP ->    VMDASAFEIFSTFPPTLYQDDTLTLQAAGLVPK-AALLLR-AR-R-AP-
                                                                                        \              /
                                                                                         AALLLR-AP-----

With our current algorithm setting, when creating the peptide cleavage graph, in end nodes of a variant bubbles are collapsed if they have the same sequence and the same outgoing nodes, and only keep the node without any variant. The graph is then turned to something like this:

VMDASAFEIFSTFPPTLYQDDTLTLQAAGLVPK-AALLLR-AR-R-AP-    ->  VMDASAFEIFSTFPPTLYQDDTLTLQAAGLVPK-AALLLR-AR-R-AP- 
                                 \              /                                         \           /
                                  AALLLR-AP-----                                           AALLLR-----

And the node AP is no longer a variant node, so then the peptide AALLLRAP is missed.

[lydiayliu]

OMG this is th edge case for bubble collapse! What are we gonna do about it?

[zhuchcn]

I initially tried to separate the reference nodes from nodes with any variants, so nodes with any variants will be collapsed together. But then it causes weird problem that one of my existing test case fails for 50% of times. And I also realized that this isn't good because synonymous mutations can also be called. For coding it's fine, but not for noncoding. So I now switched to only separate those ones with INDELs. It seems to work now. Will open a PR shortly.

[lydiayliu]

So basically ignoring the INDELs in buuble collapse, treating everything else still the same