Examples 100%

This is a manual analysis of some of the Alphafold2 models that got a WHAT IF LR-score of ~100 (that is the worst LR-score that WHAT IF can give).
Again, the LR-score tells nothing about the protein analysed itself. The score only warns when a protein structure would be a useless template for a homology modelling experiment.
Very correct natively unfolded structures, very small structures, structures that consist mainly of one very long helix, etc, all get a high LR-score despite normally being the best representation of the protein at hand.

Apology

Sorry for the shitty graphics; but remember, you got these pages for free.

AF-P0DO92-F1-model_v1.pdb

COMPND   2 MOLECULE: PUTATIVE PROTEIN T-ENOL;

MASTPMGNEGEKKSSWPSQAAPSLRGGPASLSRSEEYLSQISAELMEEALCTACCHLNPV
PIKKKQSQDQATQISKRAFFTKT

This is the sequence of a pseudo gene.

AF-P0DTW1-F1-model_v1.pdb

COMPND   2 MOLECULE: G ANTIGEN 1;

MSWRGRSTYYWPRPRRYVQPPEMIGPMRPEQFSDEVEPATPEEGEPATQRQDPAAAQEGE
DEGASAGQGPKPEADSQEQGHPQTGCECEDGPDGQEMDPPNPEEVKTPEEGEGQSQC

IUPRED2A predicts this protein to be natively unfolded over nearly the full length.

AF-P35321-F1-model_v1.pdb

COMPND   2 MOLECULE: CORNIFIN-A;

Small proline-rich protein IA (SPR-IA)

MNSQQQKQPCTPPPQPQQQQVKQPCQPPPQEPCIPKTKEPCHPKVPEPCHPKVPEPCQPK
VPEPCQPKVPEPCPSTVTPAPAQQKTKQK

SwissProt talks about repeats in the sequence. Normally, a repeat in a sequence means there either in no rigid 3D structure, or a structure with repeats. In case of the latter, no such structure exists in the PDB yet. Which means that homology modelling is not possible, not even if we would get a thousand times more sensitive BLAST. This should thus be modelled ab initio.

AF-P49006-F1-model_v1.pdb

COMPND   2 MOLECULE: MARCKS-RELATED PROTEIN;

MGSQSSKAPRGDVTAEEAAGASPAKANGQENGHVKSNGDLSPKGEGESPPVNGTDEAAGA
TGDAIEPAPPSQGAEAKGEVPPKETPKKKKKFSFKKPFKLSGLSFKRNRKEGGGDSSASS
PTEEEQEQGEIGACSDEGTAQEGKAAATPESQEPQAKGAEASAASEEEAGPQATEPSTPS
GPESGPTPASAEQNE

SwissProt tells us: Macrophage myristoylated alanine-rich C kinase substrate (Mac-MARCKS; MacMARCKS) that does something with Actin. And it has a dozen or so phosphoserines and phosphothreonines.

AF-Q4V321-F1-model_v1.pdb

COMPND   2 MOLECULE: G ANTIGEN 13;

MSWRGRSTYYWPRPRRYVEPPEMIGPMRPEQFSDEVEPATPEEGEPATQRQDPAAAQEGE
DEGASAGQGPKPEADSQEQGHPQTGCECEDGPDGQEMDPPNPEEVKTPEEGEKQSQC

Looks like I totally accedentally picked a homolog of one of the examples above. Funny to see that it then also makes the same funny loop.

AF-Q6P4E1-F1-model_v1.pdb

COMPND   2 MOLECULE: PROTEIN GOLM2;

MVGFGANRRAGRLPSLVLVVLLVVIVVLAFNYWSISSRHVLLQEEVAELQGQVQRTEVAR
GRLEKRNSDLLLLVDTHKKQIDQKEADYGRLSSRLQAREGLGKRCEDDKVKLQNNISYQM
ADIHHLKEQLAELRQEFLRQEDQLQDYRKNNTYLVKRLEYESFQCGQQMKELRAQHEENI
KKLADQFLEEQKQETQKIQSNDGKELDINNQVVPKNIPKVAENVADKNEEPSSNHIPHGK
EQIKRGGDAGMPGIEENDLAKVDDLPPALRKPPISVSQHESHQAISHLPTGQPLSPNMPP
DSHINHNGNPGTSKQNPSSPLQRLIPGSNLDSEPRIQTDILKQATKDRVSDFHKLKQSRF
FDENESPVDPQHGSKLADYNGDDGNVGEYEADKQAELAYNEEEDGDGGEEDVQDDEEREL
QMDPADYGKQHFNDVL

A large part of this looks, according to BLAST, like 4TQL computationally designed three helix bundle (DE NOVO PROTEIN 11-JUN-14 4TQL ); three helix bundle albeit with only 21% sequence identity. So, classical modelling probably would have fared better on this one than Alpha fold.

AF-Q7Z2V1-F1-model_v1.pdb

COMPND   2 MOLECULE: PROTEIN TNT;

MSLVPGQHCSPSHTRLHLTSPITMGTEPATQNTEFSKGSLIYGVTSPQRGHSQHSEASQG
PLSLDKPLQLPPIFLEGEKGESSVQNEQEGEPSLQSPSLELQSPAWPRHAGVAQEPLKVS
SSYLSDTQSSESHVSSVQHPRPEEGSHASLSSGYAGDKEGSDISLVGSHRRVRLNRRLNT
QAASNQTSQLGSIDPPSSLKSRLTGPAHSTKQTGGKE

This is a DUF (Domain of Unknown Function) in PFAM. IUPRED2A predicts this protein to contain a significant fraction of natively unfolded stretches. There is vague structural similarity between the C-terminal half of this thing and some Myosin (for which no findable structure info exists).

AF-Q9BRU2-F1-model_v1.pdb

COMPND   2 MOLECULE: TRANSCRIPTION ELONGATION FACTOR A PROTEIN-LIKE 7;

MQKPCKENEGKPKCSVPKREEKRPYGEFERQQTEGNFRQRLLQSLEEFKEDIDYRHFKDE
EMTREGDEMERCLEEIRGLRKKFRALHSNHRHSRDRPYPI

Some similarity is found by BLAST with 1WDZ (insulin receptor substrate p53). This is a thing with many big helices. It seems the secondary structure prediction capacity of Alphafold works well :-)

PDB file 1WDZ.

AF-Q96EE4-F1-model_v1.pdb

COMPND   2 MOLECULE: COILED-COIL DOMAIN-CONTAINING PROTEIN 126;

MFFTISRKNMSQKLSLLLLVFGLIWGLMLLHYTFQQPRHQSSVKLREQILDLSKRYVKAL
AEENKNTVDVENGASMAGYADLKRTIAVLLDDILQRLVKLENKVDYIVVNGSAANTTNGT
SGNLVPVTTNKRTNVSGSIR

In SwissProt I also find: "sequence caution Sequence=AAQ96871.1; Type=Erroneous gene model prediction; Evidence={ECO:0000305};"

AF-P60411-F1-model_v1.pdb

COMPND   2 MOLECULE: KERATIN-ASSOCIATED PROTEIN 10-9;

MAASTMSIRSSAYSDSWQVDDCPESCCEPPCCATSCCAPAPCLTLVCTPVSRVSSPCCQV
TCEPSPCQSGCTSSCTPSCCQQSSCQPAYCTSSPCQQACCVPVCCKPVCCVPVCCGASSC
CQQSSYQPACCASSSCQPACCVPVCCKPVCCAPTCSEDSYSCCQQSSCQPACCTSSPCQQ
SYCVPVCCKPVCCKPICCVPVCSGASSLCCQQSGCQPACCTTSCCRPSSSVSLLCRPVCR
PACCVPVSSCCAPTSSRQPSYCRQASCVSLLCRPVCSRPACYSFSSGQKSSC

This is (part of) some keratin associated protein. It holds very many short repeats. The lower right corner looks a bit compact, but that is only because this is a 2D projection, there are no contacts there.

AF-Q9HAH7-F1-model_v1.pdb

COMPND   2 MOLECULE: PROBABLE FIBROSIN-1;

MFEKYPGKMEGLFRHNPYTAFPPAVPGLPPGLPPAVSFGSLQGAFQPKSTNPELPPRLGP
VPSGLSQKGTQIPDHFRPPLRKPGKWCAMHVRVAYMILRHQEKMKGDSHKLDFRNDLLPC
LPGPYGALPPGQELSHPASLFTATGAVHAAANPFTAAPGAHGPFLSPSTHIDPFGRPTSF
ASLAALSNGAFGGLGSPTFNSGAVFAQKESPGAPPAFASPPDPWGRLHRSPLTFPAWVRP
PEAARTPGSDKERPVERREPSITKEEKDRDLPFSRPQLRVSPATPKARAGEEGPRPTKES
VRVKEERKEEAAAAAAAAAAAAAAAAAAATGPQGLHLLFERPRPPPFLGPSPPDRCAGFL
EPTWLAAPPRLARPPRFYEAGEELTGPGAVAAARLYGLEPAHPLLYSRLAPPPPPAAAPG
TPHLLSKTPPGALLGAPPPLVPAPRPSSPPRGPGPARADR

Holds (according to Swissprot) several low complexity motifs and large disordered regions. The center looks a bit compact, but that is only because this is a 2D projection, there are no contacts there.

AF-A1L429-F1-model_v1.pdb

COMPND   2 MOLECULE: G ANTIGEN 12B/C/D/E;

MSWRGRSTYYWPRPRRYVQPPEMIGPMRPEQFSDEVEPATPEEGEPATQCQDPAAAQEGE
DEGASAGQGPKPEAHSQEQGHPQTGCECEDGPDGQEMDPPNPEEVKTPEEGEKQSQC

Swissprot call the full protein disordered with as evidence MobiDB-lite.