nvalchemi.csp.chem.build_contact_distance_matrix#

nvalchemi.csp.chem.build_contact_distance_matrix(molecules, *, device='cpu')[source]#

Build intermolecular contact cutoffs used to measure overlap.

Parameters:
  • molecules (sequence of rdkit.Chem.rdchem.Mol) – Nonempty connected components in formula-unit order. Atom order is preserved, and no conformers are needed.

  • device (torch.device or str, default="cpu") – Device for the returned matrix; contact rules are evaluated on CPU.

Returns:

Owned symmetric contact cutoffs in angstroms. Entry [i, j] defines overlap between atoms i and j in different molecular copies; distances within one rigid molecule are not measured against this matrix. OverlapReliefPacker may accept residual overlap up to OverlapReliefConfig.overlap_tolerance.

Return type:

torch.Tensor, shape [A, A], dtype torch.float32

Raises:
  • OptionalDependencyError – If RDKit is unavailable. Install nvalchemi-toolkit[rdkit].

  • TypeError – If molecules is not a sequence of RDKit molecules.

  • ValueError – If the sequence is empty, a molecule is disconnected, or an element has no positive RDKit fallback van der Waals radius.

Notes

Defaults begin with van der Waals radius sums and apply the strongest symmetric donor/acceptor reduction, capped at 0.90 angstroms. The rules include hydrogen, halogen, chalcogen, phosphorus, silicon, and boron cases. Known element radii are built in; using RDKit’s periodic-table fallback emits a RuntimeWarning. These experimental chemistry rules and values may change in a future release.

Examples

This methanol input has carbon and oxygen atoms; its implicit hydrogens have no separate rows in the matrix. The C-O entry applies to atoms in different copies, not to the C-O bond inside one methanol molecule.

>>> from rdkit import Chem
>>> molecule = Chem.MolFromSmiles("CO")
>>> contacts = build_contact_distance_matrix([molecule])
>>> tuple(contacts.shape)
(2, 2)
>>> round(float(contacts[0, 1]), 2)
3.27