[phenixbb] Re: .cif file from phenix for AF3 as template

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