Exploring peptide-likeness of active molecules using 2D fingerprint methods

Similarity searching for peptide-like small molecules is a difficult task because the amide backbone shared by these molecules tends to mask features that determine biological activity. We have investigated 2D fingerprints for their ability to differentiate between peptide-like molecules having different activity or to facilitate a peptidomimetic transition from molecules with strong peptide character to compounds having little or none. For these purposes, different compound activity classes were assembled consisting of molecules having strong, moderate, and weak peptide character. For the quantification of peptide character, a "peptide flavor" index was introduced. In systematic search calculations, an encouraging finding has been that most of the investigated 2D fingerprints were capable of distinguishing between peptide-like molecules having different activities. However, only two fingerprints of different design also displayed a strong tendency to detect molecules with decreasing peptide character. One of these search tools is a recently introduced property descriptor-based fingerprint that showed two additional advantages: its flexible design could be adjusted to increasingly recover molecules with little peptide-likeness, and in addition, its search performance was not affected by differences in molecular size.