0
$\begingroup$

I used AlphaFold and Swiss-Model to get the structure of a protein (KiSS1R) and then compare them. The structures don't match and I was wondering why the softwares produce models that are different in some regions?

Thankful for all help!

$\endgroup$
2
$\begingroup$

Superposition

https://michelanglo.sgc.ox.ac.uk/data/d6e427bb-a0cb-4a65-b988-b08237f5054a

SwissModel

https://swissmodel.expasy.org/repository/uniprot/Q969F8

SwissModel models are threaded against a template. In this case the homologue has 31% identity, which isn't great. The template is 6TP3 which has a rather high R_free of 0.32 and it's solved at 3 Å, but it's a membrane structure so I personally would forgive it —others may not. SwissModel has switched primary metric to its QMEANDisCo, which is present as global (0.6 in this case) and per residue (saved in the PDB file as the temperature factor column), which is roughly similar conceptually to AlphaFold2's pLDDT (vide infra), i.e. it tells you what resides to trust. 0.6 not great, not terrible. A nice thing in SwissModel is that, by virtue of being threaded, it's easy to steal ligands and other protein and it shows models of oligomers.

AlphaFold2

https://alphafold.ebi.ac.uk/entry/Q969F8

AlphaFold2 is by far the best modelling algorithm as see in the latest CASP competition (#14). However it natively doesn't do oligomers —ColabFold notebook can.

AlphaFold2 gives a per residue score (plDDT). AlphaFold2 models the whole sequence, which results in some nasty "spaghetti loops": these have really bad scores, so need to be ignored. These are not trash, but are likely unstructured in isolation, but bind other protein etc. (more info in this blog post). This structure has N- and C-terminus loops that need to be ignored. The N in particular is the signal that would be cleaved off —to get the membrane embedding dots, one can use the OPM server.

Choice

The difference in the two models only involves low certainly regions and the support for the SwissModel is problematic. So the AlphaFold2 model is better, but requires removal of the C- and N-terminal loops.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.