protein structure prediction

homology modelling

trying to find what domains of a problem protein align well with chunks of other proteins with experimental structures available

disadvantage of homology modelling

overall fold is roughly correct but potential for many inaccuracies in the position of side chains

Phyre2, if protein does not contain domains with structures available, frankenstein approach produces no or bad results

alpha fold

relies on protein sequences and structures to calculate potential residue substitutions and distances between residues

spots which residues evolve together as a consequence of structural constraints

how does alpha fold predict its own performance

pLDTT - 0 is rubbish prediction, 100 is very confident prediction

PAE - predicted alignment error, prediction of orientation error between different domains

alpha fold disadvantages

less likely to address problems where a dimer is formed by exchanging a domain (domain swap)

does not do complexes

does not consider other biomolecules (glycosylation etc)

does not predict effect of disruptive mutations, it ignores them

produces structures that resemble bound structures where side chains point towards an inexistent ligand

does not know how to model disordered regions