Molecular Descriptors

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

Audit 0.4.1 · source sha c8c1fc3a2b13 · 19 Oct 2026Report an issue or request a re-scan

• PublisherGPTomics GitHub organizationOwner of the repository this skill is read from
• Source folderGPTomics/bioSkills/chemoinformatics/molecular-descriptorsThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdictc8c1fc3a2b13A new audit runs when the source changes
• Repository statusArchivedLast push 65 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Calculates molecular fingerprints (ECFP/Morgan, FCFP, MACCS, RDKit, AtomPair, TopologicalTorsion, Avalon, MAP4, MHFP6) and physicochemical descriptors (Lipinski, QED, TPSA, Crippen LogP, 3D shape) with explicit choice…

---
name: bio-molecular-descriptors
description: Calculates molecular fingerprints (ECFP/Morgan, FCFP, MACCS, RDKit, AtomPair, TopologicalTorsion, Avalon, MAP4, MHFP6) and physicochemical descriptors (Lipinski, QED, TPSA, Crippen LogP, 3D shape) with explicit choice tables, bit vs count semantics, and partial-charge model selection. Use when featurizing molecules for similarity, QSAR, virtual screening, or ML, or selecting the correct fingerprint for a chemotype-aware task.
tool_type: python
primary_tool: RDKit
---

## Version Compatibility

Reference examples tested with: RDKit 2024.09+, numpy 1.26+, pandas 2.2+, map4 1.1+ (MAP4), mhfp 1.9+. Use `mapchiral` separately when the stereochemistry-aware MAP4C fingerprint is intended.

Before using code patterns, verify installed versions match. If versions differ:
- Python: `pip show <package>` then `help(module.function)` to check signatures

If code throws ImportError, AttributeError, or TypeError, introspect the installed
package and adapt the example to match the actual API rather than retrying.

# Molecular Descriptors

Featurize molecules for similarity search, QSAR, virtual screening, or ML. Fingerprint performance is **dataset- and objective-dependent**: ECFP4 is a strong drug-like baseline, atom-pair and topological-torsion fingerprints expose longer-range topology, MAP4/MHFP6 target broader chemical-space searches, and 3D conformer-based descriptors are needed when shape and stereochemistry matter.

For canonicalization before featurization, see `chemoinformatics/molecular-standardization`. For 3D-only descriptors, see `chemoinformatics/conformer-generation`.

## Fingerprint Taxonomy

| Fingerprint | Type | Radius/Path | Bits | Use case | Fails when |
|-------------|------|-------------|------|----------|------------|
| Morgan (ECFP) | Circular | r=2 (ECFP4), r=3 (ECFP6) | 2048 typical | Drug-like similarity, ML default | Loses long-range topology; bit collisions at low nBits |
| FCFP | Functional Morgan | r=2 default | 2048 | Pharmacophore-aware similarity | Same caveats as ECFP; less specific |
| MACCS | Substructure key | 166 fixed bits | 167 | Quick fingerprint, drug-likeness | Too sparse for large diverse libraries |
| RDKit FP | Path/subgraph-based | paths and branched subgraphs up to 7 bonds by default | 2048 | RDKit-native ECFP alternative | Drug-like only; not optimal for scaffold hopping |
| AtomPair | Pair + topological distance | All atom pairs | 2048 | Long-range topological similarity | Slower than ECFP; harder to interpret |
| TopologicalTorsion | 4-atom torsion | All TT | 2048 | Path-pattern similarity | Like AP, slower than ECFP |
| Avalon | Substructure + atom pairs | Mixed | 512/1024 | Fast similarity | Less standard; older |
| MAP4 (MinHashed atom-pair) | MinHash atom-pair | r=1,2 | 1024/2048 | Biological + metabolite diversity | `map4` library required; slower hash |
| MHFP6 (MinHash) | MinHash ECFP-like | r=3 (diam 6) | 2048 | Large-library nearest-neighbor with a compatible MinHash/LSH index | Different distance semantics from folded-bit Tanimoto |
| Pharm2D | 2D pharmacophore | feature pairs/triplets | sparse | Pharmacophore search | Sparse, slower |

**Decision:** For drug-like similarity ranking, start with **ECFP4 2048 bit** because it is fast and well characterized. MHFP6 outperformed ECFP4 for analog recovery in the benchmark reported by Probst and Reymond (2018), making it a candidate for large, diverse libraries. For scaffold hopping, benchmark ECFP4, AtomPair, TopologicalTorsion, and pharmacophore fingerprints on target-relevant actives and decoys; published comparisons do not support a universal AtomPair advantage (Gardiner et al. 2011; Riniker & Landrum 2013).

## Bit vs Count Vectors

| Form | Use | Library impact |
|------|-----|----------------|
| Bit (0/1) | Tanimoto similarity, BulkTanimotoSimilarity, RDKit fingerprint folding | Standard for similarity |
| Count (integer) | Some ML methods, RF on counts, neural fingerprints | Loses bit-level fast operations; richer signal |
| Sparse (dict) | Direct chemical interpretation (which fragments at which atoms) | Use for SHAP / atomic attribution |

```python
from rdkit import Chem
from rdkit.Chem import rdFingerprintGenerator

mol = Chem.MolFromSmiles('CCO')

morgan = rdFingerprintGenerator.GetMorganGenerator(radius=2, fpSize=2048)
ecfp4_bit = morgan.GetFingerprint(mol)
ecfp4_count = morgan.GetCountFingerprint(mol)
ecfp4_sparse = morgan.GetSparseCountFingerprint(mol)
```

## Morgan / ECFP Radius Math

ECFP-X notation: X is the **diameter** in bonds. RDKit's `radius` parameter is half of X.

| Notation | RDKit radius | Diameter | Captures |
|----------|--------------|----------|----------|
| ECFP0 | 0 | 0 | Atom identity only |
| ECFP2 | 1 | 2 | Atom + immediate neighbors |
| ECFP4 | 2 | 4 | Atom + 2-bond environment |
| ECFP6 | 3 | 6 | Atom + 3-bond environment |

**Trade-off:** Larger radius captures more specific local environments but increases collisions at fixed `nBits`. **ECFP4 2048** is a common baseline (Rogers & Hahn 2010; Wu et al. 2018). O'Boyle and Sayle (2016) showed that increasing folded-fingerprint length can improve virtual-screening performance, but they did not establish a universal 4096-bit setting or a 1-5% collision rate. Measure collision occupancy and model performance for the dataset; increase `nBits` or use an unhashed sparse representation when needed.

## FCFP vs ECFP

FCFP (Functional-Class) uses RDKit's Morgan feature invariants (donor, acceptor, aromatic, halogen, basic, and acidic) instead of atom identity. Hydrophobe is a family in `BaseFeatures.fdef`, but it is not one of the default Morgan feature-invariant classes. FCFP trades atom-specificity for functional-equivalence.

```python
ecfp_generator = rdFingerprintGenerator.GetMorganGenerator(radius=2, fpSize=2048)
feature_invariants = rdFingerprintGenerator.GetMorganFeatureAtomInvGen()
fcfp_generator = rdFingerprintGenerator.GetMorganGenerator(
    radius=2, fpSize=

Excerpt, first 6,000 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

pip install rdkit
pip install numpy pandas

git clone https://github.com/GPTomics/bioSkills
git -C bioSkills checkout c8c1fc3a2b13

Works in

1. 2026-10-19
   ✓SAFE B 89 / 100

   first audit · source c8c1fc3a2b13

Questions about Molecular Descriptors

• github.com/GPTomics/bioSkills/tree/main/chemoinformatics/molecul — the audited source at commit c8c1fc3a2b13 and every finding aboveObserved · 19 Oct 2026

Spotted something wrong? Report a correction · Methodology