Dear Georg, Have you tried doing Molecular replacement with only the hints that you have got from Mass Spec using AF models? That would give you at least which ones could fit. The most reasonable way to me is to firstly find what is this extra density. Then I would use your 5 chains as a model for MR and then build the extra density with the protein that fits your density when not running your known sample. Best wishes ______________________________________________________ Rafael Marques da Silva PhD Student – Structural Biology University of Leicester Mestre em Física Biomolecular Universidade de São Paulo Bacharel em Ciências Biológicas Universidade Federal de São Carlos phone: +44 07861 273773 "A sorte acompanha uma mente bem treinada" ________________________________________________ ________________________________ De: Oleg Sobolev Enviado: terça-feira, 28 de outubro de 2025 19:12 Para: Georg Mlynek Cc: [email protected] Assunto: [phenixbb] Re: .cif file from phenix for AF3 as template Hi Georg, The file from PDB seems to have only 2 extra columns: _atom_site.auth_atom_id _atom_site.auth_comp_id compared to Phenix output. "Processed" mmCIF file lacks only one column compared to Phenix: _atom_site.pdbx_formal_charge. So the atom_site loop is very much similar. Does AF3 relies on these 2 items for atom names and residue names respectively? Seems unlikely since they don't output these columns from their software. It would be great to confirm or disprove with AF3 developers. I would also like to see the whole files. The error implies rather simple mistake as pointed out in your issue here: https://github.com/google-deepmind/alphafold3/issues/546 Does it really have the extra item? It would be great to get the feedback from AF3 developers as to why their library does not like Phenix mmCIF. Best regards, Oleg Sobolev. On Tue, Oct 28, 2025 at 6:51 AM Georg Mlynek wrote:

Dear Phenix community, I have solved the structure of protein complex consisting of 6 chains. We have the structure with 5 chains now and still see positive electron density and mass spec analysis gave us some possible candidates which we would like to model with AF3. However just using MSA's AF3 is not placing two chains correctly, therefore I want to give the structure templates for the 5 chains. However using the proposed way to create the files *https://github.com/google-deepmind/alphafold3/issues/462 * I get an error: ValueError: INVALID_ARGUMENT: The number of values (1958407) in a loop is not a multiple of the number of the loop's columns (19)

However it works if I use 5Y2E.cif from PDB.

Comparing the cif files I see a difference. from phenix: loop_ _atom_site.group_PDB _atom_site.id _atom_site.label_atom_id _atom_site.label_alt_id _atom_site.label_comp_id _atom_site.auth_asym_id _atom_site.auth_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.type_symbol _atom_site.pdbx_formal_charge _atom_site.label_asym_id _atom_site.label_entity_id _atom_site.label_seq_id _atom_site.pdbx_PDB_model_num

from 5Y2E 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.pdbx_formal_charge _atom_site.auth_seq_id _atom_site.auth_comp_id _atom_site.auth_asym_id _atom_site.auth_atom_id _atom_site.pdbx_PDB_model_num

After processing 5Y2E 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

Did anybody encountered the same problem and has a workaround?

Additionally I would like to ask if anybody knows which entries of the cif file header are really needed for AF3 to run? I would like to make a google colab which does the processing, without relying on the structure module from AF3 code (from alphafold3 import structure). There is a Python Package with the same name "alphafold3" on PyPi but this does not have a structure subpackage.

Many thanks, br Georg. _______________________________________________ phenixbb mailing list -- [email protected] To unsubscribe send an email to [email protected] Unsubscribe: phenixbb-leave@%(host_name)s

phenixbb mailing list -- [email protected] To unsubscribe send an email to [email protected] Unsubscribe: phenixbb-leave@%(host_name)s

What happens if you remove the “data_comp_SF4” items from the Phenix mmCIF file (ie. The last part of the file after the coordinate entries)? This item is at the end of the file to provide the PDB with information about the non-standard residues in the structure. It is possible that the alphafold code is unable to handle this (and it may be reading the data_comp_SF4 item as part of the preceding records).

On Oct 29, 2025, at 2:37 AM, Georg Mlynek wrote:

Dear Oleg, many thanks.

Fearing that it might be a problem that we are not running the latest nightly on our server I run 7B04.cif from PDB in phenix.refine latest nightly build 5861; W11.

Running the template processing in AF3 gives an error for the phenix refined .cif but not the one from PDB.

P.S.: Running my file on PDB-REDO outputs the file in a format which gives no error in af3 processing.

Many thanks, br Georg.

Am 28.10.2025 um 20:12 schrieb Oleg Sobolev:

...

Hi Georg,

The file from PDB seems to have only 2 extra columns: _atom_site.auth_atom_id _atom_site.auth_comp_id compared to Phenix output.

"Processed" mmCIF file lacks only one column compared to Phenix: _atom_site.pdbx_formal_charge.

So the atom_site loop is very much similar. Does AF3 relies on these 2 items for atom names and residue names respectively? Seems unlikely since they don't output these columns from their software. It would be great to confirm or disprove with AF3 developers.

I would also like to see the whole files. The error implies rather simple mistake as pointed out in your issue here: https://github.com/google-deepmind/alphafold3/issues/546 Does it really have the extra item?

It would be great to get the feedback from AF3 developers as to why their library does not like Phenix mmCIF.

Best regards, Oleg Sobolev.

On Tue, Oct 28, 2025 at 6:51 AM Georg Mlynek wrote:

Dear Phenix community, I have solved the structure of protein complex consisting of 6 chains. We have the structure with 5 chains now and still see positive electron density and mass spec analysis gave us some possible candidates which we would like to model with AF3. However just using MSA's AF3 is not placing two chains correctly, therefore I want to give the structure templates for the 5 chains. However using the proposed way to create the files *https://github.com/google-deepmind/alphafold3/issues/462 * I get an error: ValueError: INVALID_ARGUMENT: The number of values (1958407) in a loop is not a multiple of the number of the loop's columns (19)

However it works if I use 5Y2E.cif from PDB.

Comparing the cif files I see a difference. from phenix: loop_ _atom_site.group_PDB _atom_site.id http://atom_site.id/ <http://atom_site.id http://atom_site.id/> _atom_site.label_atom_id _atom_site.label_alt_id _atom_site.label_comp_id _atom_site.auth_asym_id _atom_site.auth_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.type_symbol _atom_site.pdbx_formal_charge _atom_site.label_asym_id _atom_site.label_entity_id _atom_site.label_seq_id _atom_site.pdbx_PDB_model_num

from 5Y2E loop_ _atom_site.group_PDB _atom_site.id http://atom_site.id/ <http://atom_site.id http://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.pdbx_formal_charge _atom_site.auth_seq_id _atom_site.auth_comp_id _atom_site.auth_asym_id _atom_site.auth_atom_id _atom_site.pdbx_PDB_model_num

After processing 5Y2E loop_ _atom_site.group_PDB _atom_site.id http://atom_site.id/ <http://atom_site.id http://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

Did anybody encountered the same problem and has a workaround?

Additionally I would like to ask if anybody knows which entries of the cif file header are really needed for AF3 to run? I would like to make a google colab which does the processing, without relying on the structure module from AF3 code (from alphafold3 import structure). There is a Python Package with the same name "alphafold3" on PyPi but this does not have a structure subpackage.

Many thanks, br Georg. _______________________________________________ phenixbb mailing list -- [email protected] mailto:[email protected] To unsubscribe send an email to [email protected] mailto:[email protected] Unsubscribe: phenixbb-leave@%(host_name)s

_______________________________________________ phenixbb mailing list -- [email protected] mailto:[email protected] To unsubscribe send an email to [email protected] mailto:[email protected] Unsubscribe: phenixbb-leave@%(host_name)s

-- Paul Adams (he/him/his) Associate Laboratory Director for Biosciences, LBL (https://biosciences.lbl.gov) Principal Investigator, Computational Crystallography Initiative, LBL (http://cci.lbl.gov) Vice President for Technology, the Joint BioEnergy Institute (http://www.jbei.org) Principal Investigator, ALS-ENABLE, Advanced Light Source (http://als-enable.lbl.gov) Laboratory Research Manager, ENIGMA Science Focus Area (http://enigma.lbl.gov) Adjunct Professor, Department of Bioengineering, UC Berkeley (http://bioeng.berkeley.edu) Member of the Graduate Group in Comparative Biochemistry, UC Berkeley (http://compbiochem.berkeley.edu) Building 91, Room 410 Building 978, Room 4126 Tel: 1-510-486-4225 http://cci.lbl.gov/paul ORCID: 0000-0001-9333-8219 Lawrence Berkeley Laboratory 1 Cyclotron Road BLDG 91R0183 Berkeley, CA 94720, USA. Executive Assistant: Michael Espinosa [ [email protected] ][ 1-510-333-6788 ] Phenix Consortium: Ashley Dawn [ [email protected] ][ 1-510-486-5455 ] --

Hello everyone, Thanks Georg for posting the issue to AF3 github repository ( https://github.com/google-deepmind/alphafold3/issues/546), developers were able to figure it out:

I understand the issue now! The problem is that the mmCIF in question has \r\n (CRLF) line endings instead of \n which we don't handle correctly because we assume \n (LF) line endings. This is a bug on our end, I am going to send a fix.

And it is fixed at this point. Best, Oleg Sobolev. On Wed, Oct 29, 2025 at 4:46 PM Georg Mlynek wrote:

Dear Paul, many thanks. Good idea. However now the processing failes without giving any error message. I also tried a version putting a # at the end of the file.

Currently I am running the phenix refined structure on PDB-Redo. Then I will compare the files from PDB and PDB-Redo which work with the one from phenix which has troubles and hopefully see a pattern.

Many thanks for the input, br Georg. Am 29.10.2025 um 16:34 schrieb Paul Adams:

What happens if you remove the “data_comp_SF4” items from the Phenix mmCIF file (ie. The last part of the file after the coordinate entries)? This item is at the end of the file to provide the PDB with information about the non-standard residues in the structure. It is possible that the alphafold code is unable to handle this (and it may be reading the data_comp_SF4 item as part of the preceding records).

On Oct 29, 2025, at 2:37 AM, Georg Mlynek wrote:

Dear Oleg, many thanks.

Fearing that it might be a problem that we are not running the latest nightly on our server I run 7B04.cif from PDB in phenix.refine latest nightly build 5861; W11.

Running the template processing in AF3 gives an error for the phenix refined .cif but not the one from PDB.

P.S.: Running my file on PDB-REDO outputs the file in a format which gives no error in af3 processing.

Many thanks, br Georg.

Am 28.10.2025 um 20:12 schrieb Oleg Sobolev:

Hi Georg,

The file from PDB seems to have only 2 extra columns: _atom_site.auth_atom_id _atom_site.auth_comp_id compared to Phenix output.

"Processed" mmCIF file lacks only one column compared to Phenix: _atom_site.pdbx_formal_charge.

So the atom_site loop is very much similar. Does AF3 relies on these 2 items for atom names and residue names respectively? Seems unlikely since they don't output these columns from their software. It would be great to confirm or disprove with AF3 developers.

I would also like to see the whole files. The error implies rather simple mistake as pointed out in your issue here: https://github.com/google-deepmind/alphafold3/issues/546 Does it really have the extra item?

It would be great to get the feedback from AF3 developers as to why their library does not like Phenix mmCIF.

Best regards, Oleg Sobolev.

On Tue, Oct 28, 2025 at 6:51 AM Georg Mlynek wrote:

Dear Phenix community, I have solved the structure of protein complex consisting of 6 chains. We have the structure with 5 chains now and still see positive electron density and mass spec analysis gave us some possible candidates which we would like to model with AF3. However just using MSA's AF3 is not placing two chains correctly, therefore I want to give the structure templates for the 5 chains. However using the proposed way to create the files *https://github.com/google-deepmind/alphafold3/issues/462 * I get an error: ValueError: INVALID_ARGUMENT: The number of values (1958407) in a loop is not a multiple of the number of the loop's columns (19)

However it works if I use 5Y2E.cif from PDB.

Comparing the cif files I see a difference. from phenix: loop_ _atom_site.group_PDB _atom_site.id http://atom_site.id _atom_site.label_atom_id _atom_site.label_alt_id _atom_site.label_comp_id _atom_site.auth_asym_id _atom_site.auth_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.type_symbol _atom_site.pdbx_formal_charge _atom_site.label_asym_id _atom_site.label_entity_id _atom_site.label_seq_id _atom_site.pdbx_PDB_model_num

from 5Y2E loop_ _atom_site.group_PDB _atom_site.id http://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.pdbx_formal_charge _atom_site.auth_seq_id _atom_site.auth_comp_id _atom_site.auth_asym_id _atom_site.auth_atom_id _atom_site.pdbx_PDB_model_num

After processing 5Y2E loop_ _atom_site.group_PDB _atom_site.id http://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

Did anybody encountered the same problem and has a workaround?

Additionally I would like to ask if anybody knows which entries of the cif file header are really needed for AF3 to run? I would like to make a google colab which does the processing, without relying on the structure module from AF3 code (from alphafold3 import structure). There is a Python Package with the same name "alphafold3" on PyPi but this does not have a structure subpackage.

Many thanks, br Georg. _______________________________________________ phenixbb mailing list -- [email protected] To unsubscribe send an email to [email protected] Unsubscribe: phenixbb-leave@%(host_name)s

_______________________________________________ phenixbb mailing list -- [email protected] To unsubscribe send an email to [email protected] Unsubscribe: phenixbb-leave@%(host_name)s

-- Paul Adams (he/him/his) Associate Laboratory Director for Biosciences, LBL ( https://biosciences.lbl.gov) Principal Investigator, Computational Crystallography Initiative, LBL ( http://cci.lbl.gov) Vice President for Technology, the Joint BioEnergy Institute ( http://www.jbei.org) Principal Investigator, ALS-ENABLE, Advanced Light Source ( http://als-enable.lbl.gov) Laboratory Research Manager, ENIGMA Science Focus Area ( http://enigma.lbl.gov) Adjunct Professor, Department of Bioengineering, UC Berkeley ( http://bioeng.berkeley.edu) Member of the Graduate Group in Comparative Biochemistry, UC Berkeley ( http://compbiochem.berkeley.edu)

Building 91, Room 410 Building 978, Room 4126 Tel: 1-510-486-4225 http://cci.lbl.gov/paul ORCID: 0000-0001-9333-8219

Lawrence Berkeley Laboratory 1 Cyclotron Road BLDG 91R0183 Berkeley, CA 94720, USA.

Executive Assistant: Michael Espinosa [ [email protected] ][ 1-510-333-6788 ] Phenix Consortium: Ashley Dawn [ [email protected] ][ 1-510-486-5455 ] --