[daylight-00]

An AtomArray passed to MPNNInferenceInput.from_atom_array_and_dict() is parsed with the parser defaults: remove_ccds and remove_waters in the input dict are ignored, and there is no way to switch off add_missing_atoms. A ligand whose label is the CCD code of a different molecule then cannot be kept as given: its atom names do not match that template, the parse raises, and the input dict offers no way around it:

# phenytoin (C15 N2 O2, 19 heavy atoms) labelled "PHT", which is phthalate in the CCD
from_atom_array_and_dict(atom_array=array, input_dict={})
# ValueError: Atom 'O1' in residue PHT (chain B, res_id 1) matches neither standard nor alternative CCD names
from_atom_array_and_dict(atom_array=array, input_dict={"add_missing_atoms": False})
# parsed ligand: the 19 atoms as given

The structure-path route already applies these overrides; only the AtomArray route dropped them, although examples/all.ipynb already passes remove_waters next to an AtomArray (with the parser's default value, so nothing showed).

• The parser fields a caller sets are forwarded to parse_atom_array() through the helper build_atom_array() now shares, so the two routes cannot drift apart again. Fields left unset keep the parser defaults.
• New optional add_missing_atoms (JSON field and --add_missing_atoms), unset by default, which still means the parser default on both routes.
• An AtomArray input behaves differently only when its dict sets remove_ccds or remove_waters, which were ignored. The defaults are untouched: they differ between the routes (remove_ccds=[] for a path, crystallization aids removed for an AtomArray), so the defaulted dict values are deliberately not forwarded.
• An AtomArray whose parse returns no assembly still falls back to structure_path.

Tested with nine new cases and the extended CLI case in models/mpnn/tests/test_inference_utils.py (CPU, AtomWorks 3.0.1). Six fail on production: the PHT case, the two ignored fields, both forwarding checks, and the new field's validation and CLI flag. The checks that pin the unchanged behavior pass on both. All of models/mpnn/tests passes (130), and the repo's pinned ruff and mypy report nothing new.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[AndrewKubaney]

This looks good to me! Thanks for making this fix!