IN SILICO
Describe molecular fingerprints before comparing them
A similarity value is meaningful only with a defined molecular representation and fingerprint configuration.
A fingerprint encodes chosen information
The RDKit book describes several fingerprint families, including Morgan fingerprints and feature based variants. These representations encode molecular information according to particular rules. A similarity calculation operates on that representation, not directly on every chemical or biological property of a compound. The fingerprint choice therefore belongs in the method description. [1]
Specify the representation and comparison
Keep the molecular processing choices, fingerprint type, radius or other relevant settings, and similarity metric in the experiment record. Preserve the source compound identifiers and the processed structures used to generate the fingerprints. If the configuration changes, treat the resulting similarity collection as a new analysis.
Inspect a small set of comparisons before screening a large library. Choose pairs that help explain what the representation considers similar and what it overlooks. A high numerical similarity may be useful for a particular retrieval task while remaining insufficient for a different scientific question. Keep the selection rule linked to the purpose of the screen.
Review the selected compounds as molecules
Use similarity to guide inspection rather than to replace it. Examine structures, relevant descriptors, and other evidence that matters to the study. When reducing a library, record which compounds were retained and why. A reproducible selection can be revised when the question changes. The goal is a transparent computational filter, not a claim that structural resemblance guarantees shared activity or the same pharmacokinetic behavior.