Scientific question

Given a query compound and a library of synthetic structures, which neighbors are structurally similar, and how do compounds spread in descriptor space?

What we would conclude with this

High Tanimoto overlap on fingerprints suggests analogs for SAR; PCA reveals scaffold clusters and outliers—useful for library design or hit expansion.

Synthetic data

120 compounds, 64-bit binary fingerprints drawn from three scaffold families; MW, logP, and TPSA-like columns derived from fingerprint density and cluster. Seed: 42. See demos/scientific-cheminformatics-similarity/data/generate.py. No RDKit—pure NumPy/Pandas.

Approach

Tanimoto coefficient on bit vectors; PCA (scikit-learn) for a 2D embedding; dual coloring by cluster and by similarity to CMP-000.

Key outputs

PCA landscape

Reproduce

cd demos/scientific-cheminformatics-similarity
python3 data/generate.py
python3 src/run.py

Dependencies: demos/requirements.txt.