Expanding ihm.LPeptideAlphabet

[gtauriello]

When handling references to large set of sequences, we ran into a couple of non-standard amino acids which should preferably be added to ihm.LPeptideAlphabet.

Here is the class as we currently use in ModelArchive:

class _LPeptideAlphabetWithXO(ihm.LPeptideAlphabet):
    """Have the default amino acid alphabet plus 'X' for unknown residues
    and 'O' as allowed non-def. AA (U already in alphabet)."""

    # extra entry added according to LPeptideAlphabet def. in
    # https://python-ihm.readthedocs.io/en/latest/_modules/ihm.html
    # and https://files.rcsb.org/view/1NTH.cif for values for 'O'.

    def __init__(self):
        """Create the alphabet."""
        super().__init__()
        self._comps['X'] = self._comps["UNK"]
        self._comps['O'] = ihm.LPeptideChemComp(
            "PYL", "O", "O", "PYRROLYSINE", "C12 H21 N3 O3"
        )
        # B/ASX, Z/GLX defined in parent class
        # J not defined in CCD? (XLE used for something else)

The non-defined 'J' (LEU/ILE AMBIGUOUS) will be an issue as soon as we remediate the ma-jd-viral model set. That one was done before we added _struct_ref to python-modelcif.

There are 5 examples in there which reference NCBI sequences containing 'J' (e.g. YP_009337833.1 for ma-jd-viral-28831) but where the model uses an 'L' instead. So to correctly handle the _struct_ref_seq_dif category for those models, we actually need a _struct_ref_seq_dif.db_mon_id for 'J' but I could not find anything in the CCD to define an ID for it.

Any suggestions on how to handle the 'J' case? (@brindakv this may need your input)

Within ModelCIF, I can of course just define a locally defined chem. comp. (i.e. by passing ccd="local" to ihm.LPeptideChemComp) but maybe there is something in the CCD which I just could not find.

[benmwebb]

This would go in python-ihm, not python-modelcif, since that's where the alphabets are defined.

I would be very reluctant to map X to UNK, since _entity_poly.pdbx_seq_one_letter_code uses (UNK) here, not X, so there is an inconsistency. But of course, no problem for you to define your own custom alphabet which does that. We could certainly add O to the alphabet though since it is an L peptide and PDB does use O for it in _entity_poly.pdbx_seq_one_letter_code (e.g. in 1nth). It looks like it's pretty rare in PDB though (RCSB says only 9 entries with it in a polymer).

I will defer to @brindakv for J.

[gtauriello]

Thanks for adding the 'O'.

Regarding X: adding it to _comps does not change how _entity_poly.pdbx_seq_one_letter_code is filled. You still get (UNK) there. The only thing it changes is that it enables us to pass "DSYVETLDXC" as sequence to modelcif.Entity instead of having to define it as ['D', 'S', 'Y', 'V', 'E', 'T', 'L', 'D', "UNK", 'C']. The written ModelCIF file looks exactly the same.

[benmwebb]

Regarding X: adding it to _comps does not change how _entity_poly.pdbx_seq_one_letter_code is filled.

Indeed. The inconsistency is that X on input gives (UNK) on output. Every other input one-letter code is unchanged on output.

The only thing it changes is that it enables us to pass "DSYVETLDXC" as sequence

Right, this is the perfect application for an Alphabet subclass, like the one you are currently using.

[gtauriello]

I see. Fair enough 👍

Then all is left for us to know how to correctly handle J (@brindakv ).