Icosahedral symmetry detection not working properly in some cases

[josemduarte]

It seems that icosahedral symmetry detection does not work properly for 1auy (a viral capsid protein with NCS ops, now properly reconstructed thanks to #114). The assemblies are built correctly: the largest assembly is one with 2 interfaces ({2,3}) and 180 subunits, i.e. an icosahedral assembly. The problem is that the symmetry detection code does not detect it being icosahedral. The symmetry detection code is based on a bunch of ad-hoc rules and needs to be better formalised.

[josemduarte]

I've added a test for this in branch issue142

[josemduarte]

I've implemented a temporary fix for this. We'll have to do something better later, I'll move the issue to the 3.1 milestone.

[sbliven]

For future reference, clusters {2,3} correspond to interfaces 31-45,70-84,100-114,61-69,85-99.

The lattice graph is quite pretty, but hard to read.

Zooming in, you can see the triangular T=3 asymmetric unit, connected by the NCS C2 axes. Both the 5-fold and 6-fold axes are induced, so every node should have the same environment.

Maybe the problem comes from the occasional duplicated edge (eg interface 61 in the lower left)?