Other examples

In this section we discuss some special cases.

AF-Q9Y2B9-F1-model_v1.pdb

COMPND   2 MOLECULE: CAMP-DEPENDENT PROTEIN KINASE INHIBITOR GAMMA;

MMEVESSYSDFISCDRTGRRNAVPDIQGDSEAVSVRKLAGDMGELALEGAEGQVEGSAPD
KEAGNQPQSSDGTTSS

Half of this sequence is found in the D-chain of PDB file 3WYG where is seems to merely sit between other domains. That part is found completely identical in this model. The rest of the model, though, probably should not exist as it is likely disordered. So, the LR score is just that, an LR-score. Half of this model is modelled 100% correct (actually modelling is a big word for obtaining it from the PDB), and the other half we probably will never know.

AF-Q9P1C3-F1-model_v1.pdb

COMPND   2 MOLECULE: PUTATIVE UNCHARACTERIZED PROTEIN PRO2829;

>AQVL
MVRPHLLKKKILGRVWWLMPVVLALWEAEVGGSLEVRSLRPAWPTW

I find the sequence VWWLMPVVLALW in this thing. That fragment can never live by itself because it is just too sticky. So, if Swissprot is not correct with "Product of a dubious CDS prediction. {ECO:0000305}" then there is another problem with this structure. In any case, just modelling it as a helix is not a good service to the scientific community; neither when the prediction is wrong, nor when it is right.