We construct fingerprint by enumerating linear and some non-linear fragments plus some information about rings. Atoms type mapping varies depending on the chain/fragment size. All this hashed into counted or non-counted vector. (counted are mostly used in prediction models)

Fingerprint properties:

• Result vector size : (1536 for default chemical similarity/distance fingerprint)
• Minimum/Maximum fragment size (chain length for linear fingerprints) : 1-6 default
• Flexible atom typing which from the combination of the following features:
• cd : element code
• hyb: hybridization
• a : aromaticity.
• h : number of connected hydrogen
• hp : number of connected hydrogen for polar atoms (0- others)
• r : ring/chain attribute
• b : number of heavy neighbors
• rs : size of the smallest ring atom belongs to (0 - for non-ring atoms)
• Flexible bond typing:
• bt : bond order
• ~r : ring/chain attribute
• ~rt : rotatable attribute
• enumeration method:
• linear chains, rings and small branched fragments (this is the default for chemical distance/similarity)
• ecfp (non-linear fragment enumeration)
• 2D ph4
• binary / counted

For the chemical distance/similarity and clustering the default fingerprint type type is 1. Length: 1536 Max linear chain length: 6 Atom typing depends on the chain length.

Custom fingerprints can be built using Descriptor function or in prediction models