Best way to make a gemmi.Model from scratch

[marinegor]

Hi everyone,

I've been implementing an mmcif reader for MDAnalysis (PR link), and was thinking also to add an MMCIFWriter in the future. That would allow to (relatively) seamlessly convert between all available formats in MDAnalysis, and mmcif, for whatever needs.

I wonder if there's a "canonical" way to do that, given e.g. we have multiple arrays with precise attribute mapping in place (see this parsing code)?

I'd be looking for something like that:

attrs: dict[str, np.ndarray] = ...
assert all(name in attrs for name in ('bfactors', 'coordinates', 'chainIDs', 'occupancies', ...))
model = gemmi.Model.from_dict(attrs)

One option that I also have in mind is to create a PDB string with built-in PDBWriter in MDAnalysis and than just read it from StringIO with gemmi, but it doesn't look canonical to me at all😁

[wojdyr]

You can make an empty Model with:

model = gemmi.Model(1)

I'm afraid there is no easy way to set coordinates, bfactors etc from arrays.
I'll keep your request in mind.
Cheers
Marcin

[wojdyr]

Actually I forgot, but we had a similar request before, and some work has been done in this direction a few month ago:
https://gemmi.readthedocs.io/en/latest/mol.html#flat-structure-representation

[marinegor]

@wojdyr that's actually cool, I'll have a look into that regarding making a model from scratch.

Another question that arises after reading that -- seems that FlatStructure is best suited for reading from bulk reads, which is exactly what we're doing with gemmi-based parser here. Currently, this ginormous map zip list whatever handles the reading, and I'd very much like to replace it with something from FlatStructure, both for speed (legacy pure python parser is faster) and readability.

However, I can't seem to find all the properties that I need in the FlatStructure representation -- could you help me out with that?

I need:

atom.altloc,  # altlocs
atom.serial,  # serials
atom.name,  # names
atom.name,  # atomtypes
# ------------------
chain.name,  # chainids
atom.element.name,  # elements
atom.charge,  # formalcharges
atom.element.weight,  # weights
# ------------------
atom.occ,  # occupancies
residue.het_flag,  # record_types
atom.b_iso,  # tempfactors
# ------------------
residue.seqid.icode,  # icodes
residue.seqid.num,  # resids
residue.name,  # resnames

# for chain in model
# for residue in chain
# for atom in residue

but I can see only

dir(flat_structure)
[
 ...
 # magic methods
 'atom_names',
 'b_iso',
 'chain_ids',
 'charge',
 'entity_ids',
 'generate_structure',
 'model_num',
 'occ',
 'pos',
 'residue_names',
 'selected',
 'subchains']

[marinegor]

In PDB, initially (in 1970's?) atom elements were encoded only in atom names.
But nowadays, they are primarily at the end of the line:

so you mean it's type_symbol from here, right? If yes, I think then indeed I should use that for atomtypes, but I now recall I couldn't find it in the atom attributes.

[wojdyr]

yes, in mmCIF it's type_symbol. BTW I added element_weights.

[marinegor]

thanks! I mean, how do I access type_symbol via gemmi and/or gemmi.FlatStructure?

[wojdyr]

thanks! I mean, how do I access type_symbol via gemmi and/or gemmi.FlatStructure?

just added FlatStructure.element_names

[marinegor]

...and it works greatly, thanks a lot :) I'll bring up the discussion to retire the legacy reader now.