Bump rdkit from 2025.9.6 to 2026.3.1

[dependabot]

Bumps rdkit from 2025.9.6 to 2026.3.1.

Release notes

Sourced from rdkit's releases.

2026_03_1 (Q1 2026) Release

Release_2026.03.1

(Changes relative to Release_2025.09.1)

Acknowledgements

(Note: I'm no longer attempting to manually curate names. If you would like to see your contribution acknowledged with your name, please set your name in GitHub)

Jakub Adamczyk, Rody Arantes, Kevin Boyd, Jessica Braun, Katharina Buchthal, Jackson Burns, Chi Cheng, David Cosgrove, Andrew Dirksen, Sergey Fedorov, Justin Gullingsrud, Tad Hurst, Lauriane Jacot-Descombes, Gareth Jones, Eisuke Kawashima, Brian Kelley, Joos Kiener, Jimmy Kromann, Phong Lam, Niels Maeder, Josh A. Mitchell, Dan Nealschneider, Yakov Pechersky, Patrick Penner, Paul Pillot, Rachael Pirie, Eliot Ragueneau, Max Rietmann, Pat Riley, Ricardo Rodriguez, Lukas Sigmund, Anton Siomchen, Raul Sofia, Matt Swain, Paolo Tosco, Ivan Tubert-Brohman, Philip Ullmann, Chris Von Bargen, Rachel Walker, Nic Zonta, 空酱, dkranthi221, EvaSnow, Pavel, stephenting22, Paul, paconius, spparel, wszqkzqk, Diogo, dehaenw

Highlights

• A significant number of known SMILES canonicalization problems were resolved.
• It's now possible to provide your own matching functions to supplement or replace the defaults when doing substructure matching.
• An experimental new implementation of shape-based alignment using Gaussians.
• Improvements to the molecular templates for 2D depiction and handling of macrocycles in the templates.

Backwards incompatible changes:

• The Dict class (and therefore all the properties interfaces) has been updated to std::string_view keys. This is transparent to the Python interfaces, but some C++ class might have to be updated.
• Simple AND queries are now merged into atoms. E.g. [C&+] now produces the the same result as [C+] when parsed as SMARTS.
• Molecules which do not have potential chiral centers or stereobonds will no longer have the "_CIPRank" atom property set by default. If you want to force the calculation of pseudo-CIP ranks, you can call Chem.ComputeAtomCIPRanks(). Note that if you just want a symmetry-aware canonical ranking of the atoms in a molecule, it is more efficient to use Chem.CanonicalRankAtoms(mol, breakTies=False).
• The behavior of H removal has changed slightly: hydrides will no longer removed by default, as this changes the global charge of the mol.
• MolToSmarts() no longer adds implicit hydrogens to atoms without queries. The one exception to this is for chiral atoms, which will still have an implicit H added if present.
• Molecules no longer have a __computedProps property by default. The property is added whenever it's needed. The property is considered private (as indicated by the leading underscores) and shouldn't be used in client code, but if your code relies on its presence in molecules, you may need to check for it and add it yourself.

... (truncated)

Changelog

Sourced from rdkit's changelog.

Release_2026.09.1

(Changes relative to Release_2026.03.1)

Acknowledgements

(Note: I'm no longer attempting to manually curate names. If you would like to see your contribution acknowledged with your name, please set your name in GitHub)

Highlights

Backwards incompatible changes:

• Since #9208, atom rings are "normalized" so that the first atom in the ring definition is the one with the lowest index, and the second one is the neighbor to the first which also has the lowest index.

Code removed in this release:

Deprecated code (to be removed in a future release):

Release_2026.03.1

(Changes relative to Release_2025.09.1)

Acknowledgements

(Note: I'm no longer attempting to manually curate names. If you would like to see your contribution acknowledged with your name, please set your name in GitHub)

Jakub Adamczyk, Rody Arantes, Kevin Boyd, Jessica Braun, Katharina Buchthal, Jackson Burns, Chi Cheng, David Cosgrove, Andrew Dirksen, Sergey Fedorov, Justin Gullingsrud, Tad Hurst, Lauriane Jacot-Descombes, Gareth Jones, Eisuke Kawashima, Brian Kelley, Joos Kiener, Jimmy Kromann, Phong Lam, Niels Maeder, Josh A. Mitchell, Dan Nealschneider, Yakov Pechersky, Patrick Penner, Paul Pillot, Rachael Pirie, Eliot Ragueneau, Max Rietmann, Pat Riley, Ricardo Rodriguez, Lukas Sigmund, Anton Siomchen, Raul Sofia, Matt Swain, Paolo Tosco, Ivan Tubert-Brohman, Philip Ullmann, Chris Von Bargen, Rachel Walker, Nic Zonta, 空酱, dkranthi221, EvaSnow, Pavel, stephenting22, Paul, paconius, spparel, wszqkzqk, Diogo, dehaenw

Highlights

• A significant number of known SMILES canonicalization problems were resolved.
• It's now possible to provide your own matching functions to supplement or replace the defaults when doing substructure matching.
• An experimental new implementation of shape-based alignment using Gaussians.
• Improvements to the molecular templates for 2D depiction and handling of macrocycles in the templates.

Backwards incompatible changes:

• The Dict class (and therefore all the properties interfaces) has been updated to std::string_view keys. This is transparent to the Python interfaces,

... (truncated)

Commits

• See full diff in compare view