AF3 pLDDT score not displaced correctly in ChimeraX

[Lukas.Bauer@…]

Hi,

I am experiencing a wired error:

For some reason the pLDDT score displayed on the AF3 server is not the same as if I downloaded the structure and displaced it in ChimeraX, by the command:

color bfactor palette 0,darkorange:49.99,darkorange:50,yellow:69.99,yellow:70,lightskyblue:89.99,lightskyblue:90,blue:100,blue

Same goes for  "color bfactor palette alphafold"

For some reason the backbone does not have the same score, however some atoms of the side chains have the “correct” score.


I don’t understand how that can be, shouldn’t the pLDDT score cover the backbone?

I wrote the AF3 DeepMind team and I was told the issue would be on the ChimeraX side, which I have a hard time to believe… I only experienced this issue with AF3 predictions, where a residue was modified by PTM…

Would you have any idea to fix this issue?

Thanks a lot.

Best regards,

Lukas

fold_2025_05_06_1_20_p42858_hd_human_s16p_p63261_actg_human_model_0.cif​

ChimeraX_pLDDT.png​

AF3_server_Screenshot.png​

[Lukas.Bauer@…]

Added by email2trac

[Lukas.Bauer@…]

Added by email2trac

[Lukas.Bauer@…]

Added by email2trac

[Elaine Meng]

Did AlphaFold3 create the CIF file?  You can clearly see within it that the bfactor field (used for pLDDT values) is different for different atoms within the same residue.  ChimeraX colors the ribbon segment by the average of the atoms in each residue whereas the atoms are colored by their specific values.  This is an excerpt from the cif file that was attached to this report.  The bfactor field is right after the occupancy field (which is 1.00 for every atom).

#
loop_
_atom_site.group_PDB
_atom_site.id
_atom_site.type_symbol
_atom_site.label_atom_id
_atom_site.label_alt_id
_atom_site.label_comp_id
_atom_site.label_asym_id
_atom_site.label_entity_id
_atom_site.label_seq_id
_atom_site.pdbx_PDB_ins_code
_atom_site.Cartn_x
_atom_site.Cartn_y
_atom_site.Cartn_z
_atom_site.occupancy
_atom_site.B_iso_or_equiv
_atom_site.auth_seq_id
_atom_site.auth_asym_id
_atom_site.pdbx_PDB_model_num
ATOM   1    N N   . MET A 1 1   ? 10.970  -12.704 27.702  1.00 32.82 1   A 1
ATOM   2    C CA  . MET A 1 1   ? 9.811   -13.602 27.623  1.00 37.72 1   A 1
ATOM   3    C C   . MET A 1 1   ? 8.531   -12.934 28.107  1.00 40.45 1   A 1
ATOM   4    O O   . MET A 1 1   ? 7.444   -13.260 27.644  1.00 37.73 1   A 1
ATOM   5    C CB  . MET A 1 1   ? 10.059  -14.863 28.451  1.00 37.89 1   A 1
ATOM   6    C CG  . MET A 1 1   ? 11.142  -15.763 27.875  1.00 34.86 1   A 1
ATOM   7    S SD  . MET A 1 1   ? 10.774  -16.308 26.198  1.00 29.69 1   A 1
ATOM   8    C CE  . MET A 1 1   ? 9.469   -17.493 26.502  1.00 25.92 1   A 1
ATOM   9    N N   . ALA A 1 2   ? 8.670   -12.012 29.046  1.00 47.72 2   A 1
ATOM   10   C CA  . ALA A 1 2   ? 7.509   -11.306 29.574  1.00 48.20 2   A 1
ATOM   11   C C   . ALA A 1 2   ? 6.803   -10.527 28.470  1.00 49.80 2   A 1
ATOM   12   O O   . ALA A 1 2   ? 5.576   -10.531 28.383  1.00 46.80 2   A 1
ATOM   13   C CB  . ALA A 1 2   ? 7.935   -10.360 30.695  1.00 44.62 2   A 1
ATOM   14   N N   . THR A 1 3   ? 7.590   -9.876  27.619  1.00 51.00 3   A 1
ATOM   15   C CA  . THR A 1 3   ? 7.032   -9.093  26.522  1.00 53.01 3   A 1
ATOM   16   C C   . THR A 1 3   ? 6.306   -10.009 25.539  1.00 56.17 3   A 1
ATOM   17   O O   . THR A 1 3   ? 5.209   -9.695  25.075  1.00 55.28 3   A 1
ATOM   18   C CB  . THR A 1 3   ? 8.134   -8.320  25.772  1.00 50.06 3   A 1
ATOM   19   O OG1 . THR A 1 3   ? 8.773   -7.407  26.669  1.00 43.33 3   A 1
[... etc. ...]

[Tom Goddard]

We discussed this multiple times on the ChimeraX mailing list in the past few years. The AlphaFold database uses single colors for whole ranges of plddt values, 70-90 is called "confident" and is light blue, and 50-70 is called "poor" and is yellow. So 69 is "poor" and 71 is "confident" and use two very different colors. This is idiotic and misleading, especially when one by luck gets a structure of many plddt values in the range 70-75 or 65-70. So ChimeraX does not use that bad AlphaFold database / DeepMind color scheme, and instead uses continuous color gradations. Here is the ChimeraX description

​https://www.cgl.ucsf.edu/chimerax/docs/user/commands/alphafold.html#colorConfidence

It would be helpful if our documentation said the colors are continuously interpolated and said it is different from the AlphaFold database discrete coloring

​https://www.ebi.ac.uk/training/online/courses/alphafold/inputs-and-outputs/evaluating-alphafolds-predicted-structures-using-confidence-scores/plddt-understanding-local-confidence