ECFP4

ECFP4, short for Extended-Connectivity Fingerprint with diameter 4, is a molecular fingerprint used to encode chemical structures for similarity searching and machine learning. It represents molecules as a fixed-length vector by encoding circular substructures around each atom up to radius 2 (diameter 4). Each atom-centered neighborhood is hashed into an identifier, and the collection of identifiers is folded into a bit vector or a count-based vector.

Generation: Starting from every atom, the method iteratively labels environments of increasing radius (0 to 2) to capture local topology, atom types, bond types, and ring membership. The resulting identifiers are then combined and folded into a fixed-length fingerprint, commonly 1024 or 2048 bits, depending on the software.

Usage: ECFP4 is widely used for similarity search with the Tanimoto coefficient and as a feature representation in QSAR, virtual screening, and machine learning models. It is valued for its balance between sensitivity to substructure and robustness across diverse chemical spaces.

Variants and software: ECFP4 is part of a family that includes ECFP6 (diameter 6) and related fingerprints sometimes referred to as Morgan fingerprints. Popular cheminformatics toolkits—such as RDKit, Open Babel, and others—offer ECFP-like generation with configurable radius and vector length, and options to include or exclude stereochemistry and isotopes.

Notes: Because fingerprints are hashed, collisions can occur, and the choice of vector length affects sparsity and performance. ECFP4 fingerprints are model-agnostic and widely adopted in chemoinformatics workflows, but performance depends on the task and data quality, and preprocessing practices can influence results.