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

review.md

48 lines · skills/datamol/references/review.md at 5f2bc03

Open on GitHub
1# Review provenance — September 30, 2026
2
3Target: Datamol 0.13.0 / RDKit 2026.03.6, Python 3.13 on macOS. This skill uses a
4local Python API; it specifies no hosted molecular-service REST endpoints. S3/GCS/HTTPS
5examples are provider-dependent paths, not verified live access or write operations.
6

Review provenance — September 30, 2026

Target: Datamol 0.13.0 / RDKit 2026.03.6, Python 3.13 on macOS. This skill uses a local Python API; it specifies no hosted molecular-service REST endpoints. S3/GCS/HTTPS examples are provider-dependent paths, not verified live access or write operations.

Raw lines 1 to 6
7## Primary sources checked
8
9- [PyPI release metadata and artifacts](https://pypi.org/project/datamol/0.13.0/):
10 release September 9, 2026, Python >=3.11. Runtime includes cloud/viz/Excel/SELFIES dependencies.
11- [Official release](https://github.com/datamol-io/datamol/releases/tag/0.13.0) and
12 [upgrade guide](https://docs.datamol.io/stable/migration.html): supported stack,
13 heterocycles descriptor-key spelling, symmetry-aware RMS, binary/JSON interchange.
14- [Release source tree](https://github.com/datamol-io/datamol/tree/0.13.0/datamol):
15 checked installed wheel source against signatures/behavior rather than relying on stale
16 prose/type hints (notably reaction return shapes and MMPA fragment strings).
17- [Molecule API](https://docs.datamol.io/stable/api/datamol.mol.html),
18 [conversions](https://docs.datamol.io/stable/api/datamol.convert.html),
19 [fingerprints](https://docs.datamol.io/stable/api/datamol.fp.html),
20 [similarity](https://docs.datamol.io/stable/api/datamol.similarity.html), and
21 [clustering](https://docs.datamol.io/stable/api/datamol.cluster.html).
22- [I/O](https://docs.datamol.io/stable/api/datamol.io.html),
23 [descriptors](https://docs.datamol.io/stable/api/datamol.descriptors.html),
24 [visualization](https://docs.datamol.io/stable/api/datamol.viz.html), and
25 [conformers](https://docs.datamol.io/stable/api/datamol.conformers.html).
26- [Scaffolds](https://docs.datamol.io/stable/api/datamol.scaffold.html),
27 [fragments](https://docs.datamol.io/stable/api/datamol.fragment.html),
28 [reactions](https://docs.datamol.io/stable/api/datamol.reactions.html), and
29 [bundled data](https://docs.datamol.io/stable/api/datamol.data.html).

Primary sources checked

Raw lines 7 to 30
31## Runtime scope
33The repository's Datamol suite executes the local reference examples and checks parsed
34failure retention, source-ID/selection correspondence, binary and JSON round trips,
35descriptor definitions, fingerprint and distance shapes, selection/cluster return contracts,
36Murcko frameworks, labelled BRICS fragments, MMPA pairs, fuzzy-scaffold DataFrames,
37reaction product groups, and the full small bundled solubility example. It verifies
38SDF/SMI/MOL/MOL2/PDB/CSV/Excel/Parquet/JSON paths, genuine PNG/SVG encodings, finite conformer
39coordinates/energies, RMS cutoff, SASA, and conformer-ID/energy correspondence after selection.
40Two-worker batches were also exercised. Parallel descriptor batching requires an explicit
41positive `batch_size` on this stack; the default None raises in joblib.
43The toy regression is an API check, not a validated benchmark. Finite conformer energy
44is not proof of minimization convergence or population weighting. Notebook widget construction
45was exercised; interactive browser rendering was not. Remote authentication, cloud writes,
46Windows FreeSASA, other Python/RDKit combinations and large-library scaling were not tested.
47Datamol explicitly disables its SASA wrapper on Windows. Source files and original molecular
48records remain necessary because a parsed table cannot recover text from invalid source records.

Runtime scope

The repository's Datamol suite executes the local reference examples and checks parsed failure retention, source-ID/selection correspondence, binary and JSON round trips, descriptor definitions, fingerprint and distance shapes, selection/cluster return contracts, Murcko frameworks, labelled BRICS fragments, MMPA pairs, fuzzy-scaffold DataFrames, reaction product groups, and the full small bundled solubility example. It verifies SDF/SMI/MOL/MOL2/PDB/CSV/Excel/Parquet/JSON paths, genuine PNG/SVG encodings, finite conformer coordinates/energies, RMS cutoff, SASA, and conformer-ID/energy correspondence after selection. Two-worker batches were also exercised. Parallel descriptor batching requires an explicit positive batch_size on this stack; the default None raises in joblib.

The toy regression is an API check, not a validated benchmark. Finite conformer energy is not proof of minimization convergence or population weighting. Notebook widget construction was exercised; interactive browser rendering was not. Remote authentication, cloud writes, Windows FreeSASA, other Python/RDKit combinations and large-library scaling were not tested. Datamol explicitly disables its SASA wrapper on Windows. Source files and original molecular records remain necessary because a parsed table cannot recover text from invalid source records.

Raw lines 31 to 48