2D pharmacophore fingerprints

There are many ways to calculate molecular similarity, and a complete discussion of the topic is beyond the scope of this chapter. The interested reader is urged to consult some of the reviews referenced here [43 8] For purposes of illustration, we will calculate molecular similarity using 2D pharmacophore fingerprints as implemented in the RDKit hbrary. Listing 6 provides the code for performing the similarity comparisons. 

The relative merits of two-, three- and four-point pharmacophore descriptors for different applications is an area of ongoing study (37, 83). Figure 5.9 shows some structurally diverse endothelin antagonists that exhibit low 2D similarity, but maintain significant overlap of their four-point pharmacophore fingerprints (37a). 

Clustering is typically used with 2D fingerprints where pairwise similarity is quantified using the Tanimoto coefficient. The sparse nature of pharmacophoric fingerprints makes them unsuitable for clustering. Clustering is a computationally expensive process and thus the size of the datasets that can be handled is limited. 

In a later study, three-point and four-point pharmacophore fingerprints have been shown to perform better than 2D Daylight fingerprints in some circumstances [48], More recently, the mini-fingerprints (MFPs), based on a small number of 2D descriptors, were specifically designed for similarity searching and were shown to have comparable performance to the 3D pharmacophores [49]. 

Some recent approaches have involved using combinations of 2D and 3D descriptors. For example, BCUT descriptors have been used in combination with four-point pharmacophores [50], and field-based methods have been combined with 2D Daylight fingerprints [47],... 

NBMs have most widely been used with sparse extended connectivity fingerprints [5, 7,12, 14, 22], Other descriptors also have been explored, such as 2D pharmacophore triplets [23] and atom types [24], Molecular fingerprints were also combined with continuous descriptors [25]. Although widely apphed naive Bayesian implementations bin continuous descriptors, the Bayesian framework naturally allows for incorporation of continuous descriptors characterized by a probabihty density, rather than a probability mass function, as in the binned case. 

NSG representations are shown to capture active compounds and their similarity relationships prior VS and following the first and second round of VS. Similarity calculations were carried out using a 2D pharmacophore-type fingerprint termed GpiDAPH3 (Molecular Operating Environment, Chemical Computing Group, Montreal, Canada). 

---