Morgan Fingerprint and Tanimoto Similarity Comparison Tool
Compare a query molecule against 2,197 approved small-molecule drugs using 2,048-bit radius-2 Morgan circular fingerprints and the Tanimoto (Jaccard) similarity coefficient. The workspace opens sildenafil in the comparison section; add tadalafil from the library to see how local atom-environment hashing measures graph overlap.
Opens sildenafil (CHEMBL192) from the shipped ChEMBL 37 library under the pinned morgan-r2-2048-described-v1 fingerprint policy. Tadalafil (CHEMBL779) is the suggested comparator; the address cannot pre-load a second molecule, so it is added in the comparison section.
Understand how circular atom-environment hashing works
A radius-2 Morgan fingerprint begins by assigning an initial integer identifier to every heavy atom based on its atomic number, degree, hydrogen count, formal charge and ring membership. In iteration one, each atom combines its identifier with the identifiers of its immediately bonded neighbours (radius 1); in iteration two, it hashes neighbourhoods extending two bonds outward (radius 2, corresponding to four-bond diameter paths). Those integer environment identifiers are then folded modulo 2,048 into a fixed-length binary bit vector.
The Tanimoto coefficient divides the count of shared on-bits by the union of on-bits present in either molecule. Because the denominator grows with molecular size, comparing a tiny fragment against a large macrocycle naturally bounds the maximum achievable Tanimoto score.
Separate 2D topological similarity from pharmacological equivalence
Use the ten nearest approved-drug neighbours in the workspace to explore structural precedent and route-of-administration context, but never infer shared target activity, selectivity or patent freedom-to-operate from a fingerprint similarity number alone.
Equations and variables
Tanimoto (Jaccard) similarity of binary bit vectors
T(A, B) = c / (a + b − c)
- a, b
- Number of on-bits set to 1 in the 2,048-bit Morgan fingerprint of molecule A and molecule B
- c
- Number of shared on-bits set to 1 in both fingerprints (|A ∩ B|)
- T(A, B)
- Dimensionless similarity coefficient ranging from 0 (no shared bits) to 1 (identical bit vectors)
Average molecular weight
MW = Σ ni Mi
- ni
- Count of atoms of element i in the represented structure
- Mi
- Average atomic mass used by the engine (g/mol)
- MW
- Molar mass of the represented structure (g/mol)
Scope and limitations
- A 2D circular fingerprint encodes local topological atom neighbourhoods up to two bonds away; two molecules can bind the same receptor pocket with similar 3D electrostatics while having a low 2D Tanimoto score.
- Conversely, an activity cliff occurs when a single methyl, halogen or stereochemical inversion barely alters the Tanimoto similarity yet abolishes biological potency.
- Tanimoto values depend strongly on fingerprint family, radius and bit length; a 0.35 score on 2,048-bit Morgan radius-2 fingerprints cannot be compared with a 0.85 MACCS keys threshold.
Sources and provenance
- RDKit descriptor implementation and versioned reference checksBSD-3-Clause
Method implementation; computed molecular properties are estimates.
- ChEMBL 37 parent structures in the shipped compound libraryCC BY-SA 3.0; share-alike
Existing library capture: 28 September 2026. Preserve attribution when redistributing its rows.
- Wildman and Crippen: atomic contributions to molecular propertiesCited method; no reproduced article data
- Ertl, Rohde and Selzer: fast calculation of molecular polar surface areaCited method; no reproduced article data
- Rogers and Hahn: extended-connectivity fingerprintsCited method; circular neighbourhood hashing algorithm
- Bajusz, Rácz and Héberger: why is Tanimoto index an appropriate choice for fingerprint-based similarity calculations?CC BY 4.0; metric comparison reference
Technical questions
What is a typical Tanimoto similarity score between two random approved drugs in Morgan radius-2 fingerprints?
In the shipped library of 2,197 approved ChEMBL small molecules using 2,048-bit radius-2 Morgan fingerprints, the median pairwise Tanimoto similarity is only 0.09, and the 99th percentile across all pairs is 0.232. A Morgan radius-2 Tanimoto above 0.35 to 0.40 already represents substantial shared local substructure.
How do Morgan radius-2 fingerprints relate to ECFP4?
Both algorithms iteratively hash atom invariants with their bonded neighbours out to a radius of two bonds (a diameter of four bonds). Because RDKit Morgan fingerprints use their own default atom invariants unless configured otherwise, they are closely analogous to ECFP4 but must be cited by their exact toolkit and policy name.
Why can two drugs with the same target, like sildenafil and tadalafil, have low 2D similarity?
Sildenafil is built on a pyrazolopyrimidinone core with a sulfonamide piperazine tail, whereas tadalafil uses a tetracyclic beta-carboline-piperazinedione scaffold with a benzodioxole ring. Their radius-2 Tanimoto is about 0.12 in an independent RDKit check, close to the library median of 0.09: the two scaffolds share few local environments even though both inhibit PDE5.
Connected workflows
- Structural edits and property comparison
Compare global fingerprint similarity with targeted single-group structural edits.
- Chemical structure and canonical SMILES
Inspect how canonical SMILES and graph topology feed into descriptor and fingerprint engines.
- US incretin listed-patent inspection
Keep chemical similarity searches strictly distinct from legal patent claim scope.