Bad mapping to PDB bioassemblies for 1a2k

[josemduarte]

There should be a {1,2} assembly (A2B2 stoichiometry, C2 symmetry) in 1a2k. As of 3.0.3, there's only {} and {1}.

[josemduarte]

This seems to be a feature and not a bug. The crystal has an unusual A2B3 stoichiometry. Engaging interface cluster 2 (interface ids 2 and 3) forms a A2B2 tetramer with C2 symmetry but the subgraph breaks the isomorphism rule for the second entity (leaves a stand-alone chain E).

Note that 5bxq is a re-modelling of the same data by same authors. The statistics are much better but the stoichiometry stays the same in their remodelling.

It seems that in the original paper the authors did check very carefully that the stoichiometry is correct in their modelling. Also they did gel filtration to find out that an A2B2 tetramer is seen in solution. So this could be a real case of co-crystal with a monomer crystallised along an A2B2 tetramer.

[lafita]

Should we then re-think or question our isomorphism rule? This is a clear example where the assumption is wrong.

Also, the PDB1 label is annotated to the {1} assembly, but the PDB1 assembly should be the {1,2} since it is tetramer plus monomer.

[josemduarte]

Absolutely this is a good case to question our isomorphism rule. But that's a difficult debate, out of the scope of this issue. We know of many other cases where it does make sense to use the isomorphism rule.

Regarding the mapping to the PDB1 assembly it does seem to be wrong in here. I'll reopen the issue changing the title.

[josemduarte]

There are quite a few other cases that mapping of assemblies EPPIC-PDB does not work. Also there are cases where the current mapping strategy takes too long (e.g. 1m4x, now avoided, see d9101c6).

Instead of matching sym ops, we need a better and more robust solution based on finding the assembly graph out of the PDB assembly and then finding the matching graph in EPPIC.