How to get the average pTM scores from an Alphafold multimer prediction
Dear Chimarax devs/users,
I'm using ChimeraX Alphafold tool to do a multimer structure prediction. I was wondering how can I determine the average pTM score for each structure prediction? I get PAE.png and plddt.png file, but not a text file or anything that tells me the average pTM. (I was hoping the ranking file would give the average score but it does not) The models come with the pdb file as well as the scores.json files, but again I don't see anything that tells me the average pTM for the structure. Clearly the pTM per residue scores are saved in my pdb file, is there a command to average them all up?
Thanks! Henry
Hi Henry, The per-residue scores in the PDB file are pLDDT values, not pTM. I don't know if pTM is included in the information that ChimeraX gets. As described in the help links below, ChimeraX shows pLDDT and PAE (predicted aligned error):
https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html#coloring https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html#pae
I hope this helps, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Department of Pharmaceutical Chemistry University of California, San Francisco
On Jan 4, 2023, at 8:27 AM, Chen, Henry via ChimeraX-users chimerax-users@cgl.ucsf.edu wrote:
Dear Chimarax devs/users,
I’m using ChimeraX Alphafold tool to do a multimer structure prediction. I was wondering how can I determine the average pTM score for each structure prediction? I get PAE.png and plddt.png file, but not a text file or anything that tells me the average pTM. (I was hoping the ranking file would give the average score but it does not) The models come with the pdb file as well as the scores.json files, but again I don’t see anything that tells me the average pTM for the structure. Clearly the pTM per residue scores are saved in my pdb file, is there a command to average them all up?
Thanks! Henry
Is there a way to get the average all the pLDDT values for the entire complex? Isn't that how the program generates its ranking?
Thanks so much, Henry
-----Original Message----- From: Elaine Meng meng@cgl.ucsf.edu Sent: Wednesday, January 4, 2023 11:40 AM To: Chen, Henry hchen38@illinois.edu Cc: chimerax-users@cgl.ucsf.edu Subject: Re: [chimerax-users] How to get the average pTM scores from an Alphafold multimer prediction
Hi Henry, The per-residue scores in the PDB file are pLDDT values, not pTM. I don't know if pTM is included in the information that ChimeraX gets. As described in the help links below, ChimeraX shows pLDDT and PAE (predicted aligned error):
<https://urldefense.com/v3/__https://rbvi.ucsf.edu/chimerax/docs/user/tools/a... > <https://urldefense.com/v3/__https://rbvi.ucsf.edu/chimerax/docs/user/tools/a... >
I hope this helps, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Department of Pharmaceutical Chemistry University of California, San Francisco
On Jan 4, 2023, at 8:27 AM, Chen, Henry via ChimeraX-users chimerax-users@cgl.ucsf.edu wrote:
Dear Chimarax devs/users,
I’m using ChimeraX Alphafold tool to do a multimer structure prediction. I was wondering how can I determine the average pTM score for each structure prediction? I get PAE.png and plddt.png file, but not a text file or anything that tells me the average pTM. (I was hoping the ranking file would give the average score but it does not) The models come with the pdb file as well as the scores.json files, but again I don’t see anything that tells me the average pTM for the structure. Clearly the pTM per residue scores are saved in my pdb file, is there a command to average them all up?
Thanks! Henry
Hi Henry,
The AlphaFold source code
https://github.com/deepmind/alphafold/blob/569eb4fea3733b979cb0442750b875759... https://github.com/deepmind/alphafold/blob/569eb4fea3733b979cb0442750b875759dd5ecc0/alphafold/model/model.py#L46
shows that the ranking for multimer predictions uses 0.8 * ptm + 0.2 * iptm
confidence_metrics['ranking_confidence'] = 0.8 * confidence_metrics['iptm'] + 0.2 * confidence_metrics['ptm']
while for monomer predictions it uses average plddt
confidence_metrics['ranking_confidence'] = np.mean( confidence_metrics['plddt']) )
The ptm, iptm and plddt values are in the score json files, for instance, this score file for a multimer prediction with 191 amino acids
~/Downloads/ChimeraX/AlphaFold/prediction_62/af191_unrelaxed_rank_1_model_5_scores.json
contains
{ "max_pae": 31.75, "pae": [[...191*191 pae values]], "plddt": [87.08, 93.55, 96.38, 95.29, 97.09, 97.85, 97.47, 97.97, 98.56, 98.23, 97.81, 98.3, 98.41, 97.97, 97.66, 98.2, 97.63, 96.64, 96.8, 95.64, 94.65, 92.39, 95.56, 97.58, 98.3, 98.57, 98.53, 98.79, 98.72, 98.77, 98.72, 98.44, 98.03, 97.95, 98.07, 97.96, 98.57, 98.61, 98.73, 98.68, 98.72, 98.67, 97.64, 96.52, 97.08, 97.62, 98.44, 98.79, 98.72, 98.79, 98.56, 98.36, 98.1, 97.43, 97.71, 96.14, 97.7, 97.76, 97.2, 97.88, 97.88, 96.67, 97.31, 97.97, 97.85, 96.7, 96.35, 96.88, 97.01, 95.29, 95.22, 95.42, 94.62, 90.54, 78.83, 65.75, 53.81, 47.67, 44.63, 42.19, 43.68, 44.19, 42.65, 39.15, 35.82, 94.59, 96.97, 98.36, 97.91, 98.02, 98.02, 98.68, 98.8, 98.76, 98.7, 98.7, 98.75, 98.59, 98.33, 98.41, 98.56, 98.05, 97.69, 98.09, 97.51, 96.95, 97.35, 97.8, 98.5, 98.78, 98.82, 98.87, 98.65, 98.35, 97.35, 96.11, 92.05, 83.29, 79.73, 73.58, 71.57, 67.75, 65.78, 61.65, 60.54, 57.76, 57.54, 51.76, 48.72, 49.38, 47.89, 49.44, 50.32, 52.88, 55.32, 60.69, 77.64, 87.56, 94.38, 96.6, 97.72, 98.55, 98.74, 98.87, 98.74, 98.0, 96.52, 95.22, 96.96, 98.41, 98.71, 98.42, 98.34, 97.5, 96.99, 94.89, 93.43, 93.7, 95.99, 96.44, 96.43, 97.62, 98.36, 97.44, 97.34, 98.49, 98.53, 97.7, 97.71, 98.45, 98.32, 97.26, 97.75, 97.98, 98.11, 96.84, 97.43, 97.96, 97.35, 94.07, 85.21, 73.09, 59.76, 52.41, 50.31, 45.78, 47.12, 43.03, 43.64, 44.21, 35.19], "ptm": 0.83, "iptm": 0.8600000143051147 }
ChimeraX does not read this scores file but you could easily read it in Python, for example from the ChimeraX Python Shell (menu Tools / General / Shell)
import json s = json.load(open('/Users/goddard/Downloads/ChimeraX/AlphaFold/prediction_62/af191_unrelaxed_rank_1_model_5_scores.json')) s['ptm']
0.83
s['iptm']
0.8600000143051147
s['plddt']
[87.08, 93.55, ...]
sum(s['plddt']) / len(s['plddt'])
88.15670157068061
The plddt values are also saved in the bfactor column of the AlphaFold PDB files. So I could also get the average plddt by averaging the bfactor value of CA atoms using the Python shell
from chimerax.core.commands import run run(session, 'select #1@CA').atoms.bfactors.mean()
88.16
The PDB format only gives two digits past the decimal point in the bfactor column so they won't match exactly the values read from the json scores file.
Tom
On Jan 4, 2023, at 9:55 AM, Chen, Henry via ChimeraX-users chimerax-users@cgl.ucsf.edu wrote:
Is there a way to get the average all the pLDDT values for the entire complex? Isn't that how the program generates its ranking?
Thanks so much, Henry
-----Original Message----- From: Elaine Meng meng@cgl.ucsf.edu Sent: Wednesday, January 4, 2023 11:40 AM To: Chen, Henry hchen38@illinois.edu Cc: chimerax-users@cgl.ucsf.edu Subject: Re: [chimerax-users] How to get the average pTM scores from an Alphafold multimer prediction
Hi Henry, The per-residue scores in the PDB file are pLDDT values, not pTM. I don't know if pTM is included in the information that ChimeraX gets. As described in the help links below, ChimeraX shows pLDDT and PAE (predicted aligned error):
<https://urldefense.com/v3/__https://rbvi.ucsf.edu/chimerax/docs/user/tools/a... > <https://urldefense.com/v3/__https://rbvi.ucsf.edu/chimerax/docs/user/tools/a... >
I hope this helps, Elaine
Elaine C. Meng, Ph.D. UCSF Chimera(X) team Department of Pharmaceutical Chemistry University of California, San Francisco
On Jan 4, 2023, at 8:27 AM, Chen, Henry via ChimeraX-users chimerax-users@cgl.ucsf.edu wrote:
Dear Chimarax devs/users,
I’m using ChimeraX Alphafold tool to do a multimer structure prediction. I was wondering how can I determine the average pTM score for each structure prediction? I get PAE.png and plddt.png file, but not a text file or anything that tells me the average pTM. (I was hoping the ranking file would give the average score but it does not) The models come with the pdb file as well as the scores.json files, but again I don’t see anything that tells me the average pTM for the structure. Clearly the pTM per residue scores are saved in my pdb file, is there a command to average them all up?
Thanks! Henry
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
participants (3)
-
Chen, Henry -
Elaine Meng -
Tom Goddard