Structure guide · 10/15/2026
When similarity scores differ: six steps to reproduce a comparison
Track molecular inputs, fingerprint settings and bit overlap before interpreting different scores.
Read the two structures
Locate ring nitrogens in gefitinib and erlotinib, then compare the substituents outside the rings. Check halogen labels and side-chain differences using both formula and drawing. A shared local motif does not establish identity of the whole molecule.
| Observation | Gefitinib | Erlotinib |
|---|---|---|
| Carbon triple bond C≡C | 0 | 1 |
| Fluorine atoms | 1 | 0 |
| Chlorine atoms | 1 | 0 |
Connect the structural observations with the fingerprint result
Gefitinib has one highlighted fluorine and one chlorine. Erlotinib has neither of these query matches and instead has one carbon triple-bond C≡C match. Read the coloured regions with the counts to distinguish the selected substituents concretely. [Gefitinib] · [Erlotinib]
There are 31 shared set bits across a union of 76, giving 0.408. The selected substituents differ while other encoded environments overlap. This is a whole-structure fingerprint calculation, not a score comparing only halogens and the triple bond.
Visual resemblance is a useful observation, but this calculation neither computes a maximum common substructure nor establishes an identical scaffold. Fingerprint overlap and a shared-scaffold claim require different evidence.
Calculated structural similarity: 0.408 · Morgan radius 2 / 2048 bits · Tanimoto
Calculation: 31 / (60 + 47 − 31) = 31 / 76 ≈ 0.408. Set-bit counts are not counts of unique atoms or functional groups.
Query definitions and reproducibility
- Carbon triple bond C≡C:
C#C - Fluorine atoms:
[F] - Chlorine atoms:
[Cl]
RDKit 2025.03.4 · Matches with identical atom sets are counted once. Colours mark atoms and bonds matching selected queries, not a proven shared scaffold or synthesis route.
- Read elemental composition from the formulas.
- Trace ring and substituent connections in each drawing.
- Compare your observations with the fingerprint score calculated under identical settings.
- Check stereochemistry and the complete InChIKey in the source records.
The score is fingerprint overlap, not a percentage of efficacy or safety. RDKit
Structure-source check: 2026-10-08 / 2026-10-08 · Editorial update and calculation check: 2026-10-16
Download source structures, queries and bit counts for this comparison (JSON)
1 · freeze the molecular inputs
Record both CIDs, complete InChIKeys, SMILES and source dates. The same name can conceal different salts, parent structures or stereochemical representations. Use the downloadable comparison to preserve the exact inputs instead of searching the names again at each step.
2 · check parsing and identity separately
Parse each input and compare its drawing with the source record. A successful parse does not prove that a valid structure belongs to the named molecule. This page uses gefitinib–erlotinib as an example; its bit counts must not be reused for other pairs.
3 · write down the fingerprint settings
The default here is a radius-2 Morgan bit fingerprint with 2048 bits and chirality not separately enabled. Radius is a neighborhood expansion parameter, not a distance of two angstroms. Changing size or switching to a count fingerprint changes the representation. Preserve the software version as well.
4 · check intersection and union
In a hypothetical example, A has 70 set bits, B has 60 and 40 overlap. The union is 90 and Tanimoto is 40/90. These illustrative values are not the measurements for the molecules below. Preserve unrounded results so display precision can be distinguished from a real calculation difference.
5 · change one condition at a time
Match the input first, then compare normalization and stereochemical treatment, fingerprint type, radius, size and implementation version. Changing everything at once hides the cause. Without rerunning a changed condition, leave the size of its effect unknown rather than inventing a sensitivity result.
6 · distinguish reproduction from validation
Obtaining the same number under the same settings checks reproducibility. It does not validate prediction of activity. Report inputs, settings, result and limitations together. A score of one without chirality-sensitive encoding still requires checking the original stereochemical record.
Sources and review
Editorial update: 10/15/2026 · Structure-source check: 2026-10-08. Calculations and cited experiments are separate evidence.
Related compounds
Editorial responsibility and corrections
Operated and edited by the independent StructureMoa project. Explanations use AI assistance; structural identity and similarity are checked against source records and calculation tests. Molecular data and clinical or regulatory evidence have separate review scopes.
- Source structures
- PubChem · CID / SMILES / InChIKey
- Calculations and automated checks
- RDKit · Morgan / SMARTS
- Disclosed specialist review
- No separate external specialist review is disclosed. Passing automated tests is not specialist review of clinical claims.
For a correction, include the page URL, molecule name or CID, the field in question and a supporting source. Corrections are checked against sources, and the affected editorial update date is changed when a correction is published. [email protected]