Guides¶
One page per capability, organised by what you are trying to do rather than by which crate the code lives in.
-
Parsing, storage order vs canonical output, chirality and double-bond geometry, explicit hydrogens.
-
What each stage does, in what order, and why the order matters.
-
SMARTS queries, match ordering, uniquification, and stereo-aware matching.
-
Applying templates, atom mapping, intramolecular reactions, and where the product count comes from.
-
Rebuilding the fragments a template discards — and reading the verdict that tells you how much to trust the answer.
-
V2000 molblocks and multi-record SDF, in 2D and 3D, and how stereochemistry survives the round trip.
-
2D coordinates and SVG/PNG/JPEG output, the two drawing styles, and what gets reported when a structure cannot be drawn well.
-
Space-filling, ball-and-stick, stick and wireframe, in CPK colours — how the viewpoint is chosen, and when it carries no information.
-
One deterministic conformer per molecule — why there is no random seed, and what the numbers that come back with it mean.
-
The per-atom and per-bond values a graph neural network reads — including Gasteiger partial charges — and why they come back as names, not one-hot vectors.
-
The columnar representation, and what it is for.
The one idea behind all of them¶
A molecule's properties are decided by the molecule, not by how someone chose to write it down.
Every layer is checked against that. Two SMILES for the same structure must sanitize to the same thing, match the same queries, and react the same way. Where an implementation choice could break it — the order neighbours happen to be stored in, which atom a template happened to be written around — the design notes say so explicitly, because those are exactly the places where bugs are invisible from the outside.