Skill Tracer
Skills › Scientific Agent Skills › rdkit › review.md

review.md

80 lines · skills/rdkit/references/review.md at 5f2bc03

Open on GitHub
1# RDKit review record
2
3Reviewed 2026-10-01 against stable RDKit **2026.03.6**, PyPI package
4`rdkit==2026.3.6`, executed on macOS ARM64 / Python 3.13. Development branches and
5pre-release APIs were not adopted. `FindPotentialStereo` and multithreaded
6suppliers remain explicitly experimental even in the stable package.
7

RDKit review record

Reviewed 2026-10-01 against stable RDKit 2026.03.6, PyPI package rdkit==2026.3.6, executed on macOS ARM64 / Python 3.13. Development branches and pre-release APIs were not adopted. FindPotentialStereo and multithreaded suppliers remain explicitly experimental even in the stable package.

Raw lines 1 to 7
8## Official sources checked
9
10- [Stable release](https://github.com/rdkit/rdkit/releases/tag/Release_2026_03_6)
11 and [PyPI metadata](https://pypi.org/pypi/rdkit/json): stable version and wheel availability.
12- [Installation](https://www.rdkit.org/docs/Install.html) and
13 [Python introduction](https://www.rdkit.org/docs/GettingStartedInPython.html):
14 binary packaging, parsing, suppliers, conformers, drawing.
15- [RDKit Book](https://www.rdkit.org/docs/RDKit_Book.html): SMARTS semantics,
16 sanitization, stereochemistry, reactions, hydrogen handling, thread caveats.
17- [2026.03.1 release notes](https://github.com/rdkit/rdkit/releases/tag/Release_2026_03_1):
18 canonical stereo/Kekulization changes, H removal and MolToSmarts behavior.
19 The standalone backwards-incompatible-changes page stops at older releases;
20 release notes are necessary for current changes.
21- [Molecular files](https://www.rdkit.org/docs/source/rdkit.Chem.rdmolfiles.html),
22 [graph operations](https://www.rdkit.org/docs/source/rdkit.Chem.rdmolops.html),
23 [molecule/atom/bond API](https://www.rdkit.org/docs/source/rdkit.Chem.rdchem.html).
24- [Fingerprint generators](https://www.rdkit.org/docs/source/rdkit.Chem.rdFingerprintGenerator.html),
25 [atom pairs](https://www.rdkit.org/docs/source/rdkit.Chem.AtomPairs.Pairs.html),
26 [torsions](https://www.rdkit.org/docs/source/rdkit.Chem.AtomPairs.Torsions.html),
27 [similarities](https://www.rdkit.org/docs/source/rdkit.DataStructs.html),
28 [Butina](https://www.rdkit.org/docs/source/rdkit.ML.Cluster.Butina.html).
29- [Descriptors](https://www.rdkit.org/docs/source/rdkit.Chem.Descriptors.html),
30 [molecular descriptors](https://www.rdkit.org/docs/source/rdkit.Chem.rdMolDescriptors.html),
31 [tagged Lipinski definitions](https://github.com/rdkit/rdkit/blob/Release_2026_03_6/rdkit/Chem/Lipinski.py).
32- [Standardization API](https://www.rdkit.org/docs/source/rdkit.Chem.MolStandardize.rdMolStandardize.html)
33 and [released implementation](https://github.com/rdkit/rdkit/blob/Release_2026_03_6/Code/GraphMol/MolStandardize/MolStandardize.cpp):
34 Cleanup retains disconnected salts; FragmentParent, Uncharger and tautomer
35 operations are separate, scientifically consequential policies.
36- [Embedding](https://www.rdkit.org/docs/source/rdkit.Chem.rdDistGeom.html),
37 [released force-field wrapper](https://github.com/rdkit/rdkit/blob/Release_2026_03_6/Code/GraphMol/ForceFieldHelpers/Wrap/rdForceFields.cpp),
38 [drawing](https://www.rdkit.org/docs/source/rdkit.Chem.Draw.html).
39- [Reactions](https://www.rdkit.org/docs/source/rdkit.Chem.rdChemReactions.html),
40 [hashes](https://www.rdkit.org/docs/source/rdkit.Chem.rdMolHash.html),
41 [scaffolds](https://www.rdkit.org/docs/source/rdkit.Chem.Scaffolds.MurckoScaffold.html),
42 [PAINS catalogue](https://www.rdkit.org/docs/source/rdkit.Chem.FilterCatalog.html).
44Installed stable docstrings/signatures supplemented web pages that omitted
45compiled-function details. These are local APIs: no hosted chemistry endpoint,
46credential, pagination, remote inference, or scientific database request is
47required by the bundled scripts.

Official sources checked

Installed stable docstrings/signatures supplemented web pages that omitted compiled-function details. These are local APIs: no hosted chemistry endpoint, credential, pagination, remote inference, or scientific database request is required by the bundled scripts.

Raw lines 8 to 48
49## Executed checks and boundaries
51The isolated per-skill suite exercises the three CLIs plus shared file reader,
52known positive/negative functional groups, invalid exclusions, source record
53identity, empty inputs, query cardinality, fingerprint dimensions/thresholds,
54enantiomers, specified versus unspecified stereo, isotope/charge/tautomer
55identity, Cleanup/fragment/uncharging distinctions, valence failure, converged
56conformers, force-field coverage, units, reactions and depiction copies.
58Supplementary tiny-molecule smoke checks covered 72 named descriptors, 12
59`rdMolDescriptors` entries, MOL/SDF/SMILES/gzip/PDB/InChI/peptide I/O, multithreaded
60source IDs, graph edits, hashes/scaffolds/features, Avalon/ErG, legacy fingerprint
61aliases, bit visualization, similarity metrics/Butina, UFF/MMFF, RMSD/alignment,
62constrained embedding, template depiction and reaction fingerprints. An attribute
63and signature audit covered the entire named module API inventory.
65Two observed behaviors deserve attention:
67- Fresh `SmilesMolSupplier` passed directly to `list()` returned no records in
68 the tested build; iteration/comprehensions worked. The bundled reader parses
69 physical SMILES lines itself and preserves their source IDs.
70- Chirality-aware 2048-bit atom-pair fingerprints collided for simple
71 enantiomers; sparse count representations distinguished them. Similarity 1
72 must never substitute for a declared chemical identity policy.
74Reference snippets with input paths, undefined molecules/indices or notebook
75objects are illustrative templates. MOL2 chemical-typing conventions and optional
76Jupyter integration are documentation/signature reviewed, not executed here.
77No performance/concurrency stress benchmark, cross-platform binary verification,
78experimental validation, pH/tautomer-population inference, or predictive model
79validation is claimed. Tiny organic examples do not establish parameter coverage
80for organometallics, unusual valences, macrocycles or proteins.

Executed checks and boundaries

The isolated per-skill suite exercises the three CLIs plus shared file reader, known positive/negative functional groups, invalid exclusions, source record identity, empty inputs, query cardinality, fingerprint dimensions/thresholds, enantiomers, specified versus unspecified stereo, isotope/charge/tautomer identity, Cleanup/fragment/uncharging distinctions, valence failure, converged conformers, force-field coverage, units, reactions and depiction copies.

Supplementary tiny-molecule smoke checks covered 72 named descriptors, 12 rdMolDescriptors entries, MOL/SDF/SMILES/gzip/PDB/InChI/peptide I/O, multithreaded source IDs, graph edits, hashes/scaffolds/features, Avalon/ErG, legacy fingerprint aliases, bit visualization, similarity metrics/Butina, UFF/MMFF, RMSD/alignment, constrained embedding, template depiction and reaction fingerprints. An attribute and signature audit covered the entire named module API inventory.

Two observed behaviors deserve attention:

  • Fresh SmilesMolSupplier passed directly to list() returned no records in the tested build; iteration/comprehensions worked. The bundled reader parses physical SMILES lines itself and preserves their source IDs.
  • Chirality-aware 2048-bit atom-pair fingerprints collided for simple enantiomers; sparse count representations distinguished them. Similarity 1 must never substitute for a declared chemical identity policy.

Reference snippets with input paths, undefined molecules/indices or notebook objects are illustrative templates. MOL2 chemical-typing conventions and optional Jupyter integration are documentation/signature reviewed, not executed here. No performance/concurrency stress benchmark, cross-platform binary verification, experimental validation, pH/tautomer-population inference, or predictive model validation is claimed. Tiny organic examples do not establish parameter coverage for organometallics, unusual valences, macrocycles or proteins.

Raw lines 49 to 80