molecular fingerprint

A molecular fingerprint is a long string of ones and zeros that encodes which structural features a molecule contains, like a checklist where each box is ticked if the molecule has that particular fragment. Just as a human fingerprint summarizes an identity in a compact pattern, a molecular fingerprint summarizes a structure in a compact bit-string.

Different fingerprint schemes choose different features. Some tick a box for each substructure from a fixed dictionary of chemical patterns. Others, such as circular fingerprints, enumerate the atom environments around each atom out to a given radius and hash them into bits. Either way, the result is a vector that computers can compare extremely fast, even across millions of molecules.

Their main use is measuring similarity: two molecules whose fingerprints share many set bits are deemed structurally alike, often quantified by a similarity coefficient. This drives similarity searching, clustering of libraries, and one route into machine learning. The caveat is that fingerprint similarity is a proxy for chemical resemblance, not for biological activity, so two molecules can look similar by fingerprint yet behave very differently, the activity-cliff problem again.