lafita
commented
9 years ago I close the issue because it seems a problem with the protein (in the benchmark it is reported as 8-fold), rather than a bug in the refinement/order detection. If any other example of this type is found the issue can be opened again.
The results of the SINGLE refiner on the PDB code 2I5I.A show inconsistent connected residues, that are too close in sequence. This results in a very short alignment and a poor subunit identification (notice, as an example, that brown and yellow residues are part of the same subunit):
This can be considered a bug, because in the alignment graph of the original alignment, before refinement, (left) we can observe that there are at least 10 groups of 7 consistent connected residues, which have to be identified by the refiner without any doubt. Instead, the residues connected in the refined alignment (left) form only 4 groups and they are not consistent, because 63-59, 62-58, etc are too close in sequence that they must be either part of a same subunit or part of none.
A solution might be to check for consistency (a minimum residue distance between connected residues) in the Refinement algorithm, because no code fragment does it. By now only this example has been found to produce non-consistent refined alignments.