ECFPn

ECFPn refers to Extended-Connectivity Fingerprints with a specified diameter n, a class of circular molecular fingerprints used to encode structural features of chemical compounds. In this scheme, n denotes the fingerprint diameter (with radius roughly equal to n/2), and common examples include ECFP4 and ECFP6, corresponding to radii of 2 and 3, respectively. The ECFPn family is a type of Morgan fingerprint, deriving from circular atom environments rather than linear substructures.

Computationally, ECFPn begins by assigning each atom a simple invariant based on properties such as atomic number, valence, and in some implementations stereochemistry or isotopes. The algorithm then iteratively expands the local environment around each atom up to radius r = n/2 by combining the current atom invariant with invariants of neighboring atoms and bonds, producing a set of environment identifiers. These identifiers are typically hashed and mapped into a fixed-length binary vector (for example 1024 or 2048 bits). Each substructure that hashes to a given position sets the corresponding bit. Some implementations also provide counts of substructure occurrences rather than a binary presence.

Commonly used variants are ECFP4 (radius 2) and ECFP6 (radius 3), though other diameters are possible. ECFPn fingerprints are widely used for similarity searching, virtual screening, and machine learning in cheminformatics, often in combination with the Tanimoto similarity measure. They are valued for capturing local topology and substructural information while remaining computationally efficient.

Strengths of ECFPn include robustness to small structural changes and effectiveness in ligand-based modeling. Limitations include potential information loss for long-range interactions beyond the chosen radius, collisions from hashing, and a lack of explicit 3D or conformational information. ECFPn is one of the most commonly used circular fingerprints and is often preferred over non-circular or purely 2D descriptors for many applications.