How to Use Chemical Library Analysis

Submit up to 100,000 small molecules as SMILES, SDF structures, or Chemical Component Dictionary codes. Chemical Library Analysis describes the collection through compounds, Bemis–Murcko scaffolds, scaffold hierarchies, fragments, and molecular similarity. The results include a library overview, chemical islands, compound assignments, scaffold and fragment support, and similarity connections.

The analysis uses 2,048-bit Morgan fingerprints with the bits from radii 2 and 3 combined. Chirality is not encoded in the similarity fingerprints. Pairwise similarities are scanned exactly for up to 2,000 compounds. Larger libraries use approximate MinHash/LSH candidate search followed by exact Tanimoto rescoring; qualifying pairs can be missed. Each compound selects up to eight strongest retrieved neighbors meeting Similarity Threshold, and the graph retains the union of these selections. A compound can consequently have more than eight retained neighbors. Chemical islands are communities of this sparse compound-similarity graph, using Cluster Resolution to control grouping. They are distinct from components of the full graph, which also connects compounds through scaffolds and fragments. A compound without a retained similarity connection forms its own island; this does not establish chemical novelty.

Scaffolds start from Bemis–Murcko ring frameworks and expand into a hierarchy. Fragment analysis combines BRICS, rotatable-bond cuts, and linker fragments. Compound Support counts submitted entries directly linked to each scaffold or fragment; Library Fraction divides this support by the number of submitted entries. Hierarchy-only scaffolds can have zero direct support. Entropy describes how support is distributed, and Gini describes scaffold concentration. All coverage and exclusivity statements are relative to the submitted library.

Canonical SMILES retain stereochemistry, but the analysis does not standardize salts, protonation states, or tautomers. Ordinary explicit hydrogens are removed for consistent connectivity comparison. Repeated canonical structures and alternate conformers remain separate compounds and affect counts, support, and grouping. Stable Compound IDs preserve input order even when names repeat. Repeated Structures counts entries beyond the first occurrence of each canonical SMILES.

The compound table reports island assignments, retained similarity-neighbor counts, and whether a compound participates in a similarity edge between islands. Islands are numbered from one. Mean Edge Similarity summarizes retained internal edges rather than all pairs; it is blank when an island has no such edge. The example molecule is one member of the island, not an optimized representative.

Download summary.csv, compounds.csv, islands.csv, scaffolds.csv, fragments.csv, similarity_edges.csv, and bridges.csv for analysis. The chemical_graph.json and chemical_graph.graphml files retain the full compound/scaffold/fragment graph; compound IDs link graph nodes to names in compounds.csv. Scaffold support and chemical islands are plotted interactively on the results page. Structural similarity and graph relationships describe the collection and do not predict activity or establish a synthetic route.

• Library overview: Review scaffold diversity, fragment support, and repeated structures together.
• Chemical families: Inspect similarity communities, isolated compounds, and links between groups.
• Traceable compounds: Preserve input order and repeated names with unique compound IDs.
• Reusable results: Download tables and graph relationships for further analysis.