Molecular descriptors

What are molecular descriptors?

Molecular descriptors turn a chemical structure into numeric features. For small molecules, these features describe size, polarity, lipophilicity, hydrogen bonding, ring systems, synthetic accessibility, and structural patterns that can be used for drug-likeness screening, QSAR modeling, similarity search, and machine-learning workflows.

ProteinIQ calculates descriptors from SMILES with RDKit. The default output includes a compact drug-discovery summary first, then RDKit's full descriptor catalog so the same result can support both quick inspection and downstream modeling.

How to use Molecular Descriptors online

Paste SMILES strings, upload a SMILES file, or fetch compounds from PubChem to calculate RDKit descriptors online. ProteinIQ returns a downloadable table with physicochemical properties, QED, SA Score, Lipinski, Veber, and Ghose rule violations, plus optional Morgan/ECFP4 and MACCS fingerprints for ML-ready feature export.

Inputs

Settings

Results

The All descriptors table starts with interpretable columns that are commonly used in medicinal chemistry, then appends RDKit descriptor columns. RDKit catalog values are rounded to four decimal places in the table for stable export and readable wide spreadsheets.

Interpreting results

Descriptor filters are fast triage tools. They identify molecules that look unusual relative to historical oral drugs, not molecules that are guaranteed to fail.

Rule violations

Zero or one violation is usually compatible with early oral-drug screening. Two or more violations need context. Natural products, macrocycles, targeted degraders, covalent ligands, and non-oral compounds often sit outside these ranges for valid reasons.

Fingerprints for ML and similarity search

Fingerprints encode molecular substructures as binary vectors. They are not descriptors in the same sense as MW or LogP; they are sparse structural features for comparing molecules or training models.

The fingerprint preset returns bitstrings rather than thousands of visible spreadsheet columns. Compact bitstring output keeps the browser table readable while preserving exportable ML features.

How Molecular Descriptors works

RDKit parses each SMILES string into a molecular graph. Descriptor functions then operate on that graph without generating 3D conformers. The calculation is fast and deterministic for large compound lists, but the output does not capture conformational strain, 3D shape, explicit solvation, metabolism, or protein-ligand interactions.

The All descriptors preset uses RDKit's descriptor catalog through Descriptors.CalcMolDescriptors. That catalog includes:

• Constitutional descriptors: Atom counts, molecular weight, exact mass, valence electrons, heteroatom counts, and ring counts.
• Topological descriptors: Chi indices, Kappa shape indices, BalabanJ, BertzCT, and related graph features.
• Surface-area descriptors: TPSA, LabuteASA, PEOE_VSA, SMR_VSA, SlogP_VSA, EState_VSA, and VSA_EState columns.
• Electronic descriptors: EState indices, partial-charge summaries, and BCUT2D columns.
• Fragment descriptors: Functional group counts such as fr_Al_OH, fr_ester, fr_amide, fr_halogen, and fr_benzene.

Invalid SMILES rows are reported with an error column instead of stopping the whole job.

When to use Molecular descriptors vs alternatives

Molecular Descriptors is the broad feature-generation tool. It is best when a workflow needs raw properties, RDKit descriptors, or fingerprints for custom filtering, QSAR, clustering, or model input.

Use a focused filter when the goal is a single decision rule rather than a wide feature table.