Cheminformatics Module

Screen a Set of Molecules Before You Simulate Them

Properties, fingerprints, similarity, substructure search, clustering and 3D conformers, then send the best candidates to quantum chemistry.

Get Started Request a Demo

Two chemists at a standing desk reviewing a monitor with a grid of skeletal molecular structures, a few highlighted

Five Tools on a Shared Molecule

Runs on the CPU of your own server. Live

Properties

Formula, molecular weight, Wildman-Crippen logP, Ertl TPSA, hydrogen-bond donors and acceptors, rotatable bonds, rings, and the Lipinski and Veber rules.

Similarity

ECFP4 (radius 2, 2048 bits) and path fingerprints with Tanimoto coefficients against any molecule in your library.

Substructure

VF2-style matching of a pattern across a whole collection.

Cluster

Butina clustering of fingerprints to pick one representative per cluster.

Conformer

3D coordinates by distance geometry, ready for a quantum chemistry optimisation.

Tested Against RDKit and Open Babel

Both programs are used only as test references in a separate environment. Nothing from them ships in the engine.

906 of 907

Checks agreeing with RDKit

597 of 600

Checks agreeing with Open Babel

Canonical SMILES identical across 25 random atom orders for 20 molecules, and across 52 stereo structures
3000 fuzzed stereo SMILES tested
2000 fuzzed malformed inputs: 1633 cleanly refused, none crashed the engine
Unusual valence gives a warning, never a refusal, so perchlorate, XeF2 and hypervalent iodine compounds are accepted
Results show the Lipinski count, the Veber rules, units, the Tanimoto definition used, and a stereo and isotope line

Known Differences, in the Open

Sulfonic acid logP: 3.117 here and in Open Babel, 0.955 in RDKit (register item SI-03)
Crippen types for O7, N14 and metals are not implemented; logp_complete is then false
3D coordinates differ up to 0.026 Å across compilers, so regression compares them at 0.05 Å
MOL and SDF (V2000 and V3000) and PDB files are read and written by the engine since 0.7.7, checked 736 of 736 against a reference toolkit; CIF is planned

A chemist at a laptop in a bright lab asking the AI assistant a question, with a molecule and suggestion chips on screen

Cheminformatics Questions

What does the cheminformatics module compute?

Canonical SMILES, formula, molecular weight (IUPAC 2021 standard atomic weights), monoisotopic mass (AME2020), Wildman-Crippen logP, Ertl TPSA, hydrogen-bond donors and acceptors, rotatable bonds, ring counts and Lipinski violations. It also does ECFP4 and path fingerprints with Tanimoto similarity, substructure search, Butina clustering, 2D depictions and 3D conformers.

How large a set can I process?

Up to 500 molecules per call, and 200 for 3D conformers. A screening set can send up to 50 molecules at once to quantum chemistry.

Does it handle stereochemistry and isotopes?

Yes for canonical SMILES: tetrahedral and double-bond marks are kept and the output does not depend on atom order, following OpenSMILES. Isotopes such as [2H] and [13C] use AME2020 masses. Formula, descriptors, fingerprints and 3D coordinates do not yet use stereo.

Are there known differences from RDKit or Open Babel?

Yes, and they are published. Sulfonic acid logP differs from RDKit (3.117 against 0.955) while Open Babel agrees with us, and Crippen types for O7, N14 and metals are not implemented, so the result sets a logp_complete flag to false.

Run Your Next Molecule Through AiMolSim

See your own molecule optimised, its frequencies computed and the checks printed beside the result. A demo takes about 30 minutes.