Similarity Methods Based on Substructures

Willett and co-workers ° have developed several novel and useful techniques in molecular similarity based on substructural fragments. These approaches are based on the frequency of occurrence of generated fragment descriptors within the molecular graph. Success of these methods has been shown in 2D and 3D matchings of chemical structure, classification of chemical data bases, as well as property estimation. 

Basak and co-workers also used K K= 1-10, 15, 20, 25) nearest neighbors of compounds in predicting properties like lipophilicity, boiling point, and mutagenicity ° of diverse data bases. For a structurally diverse set of 76 compounds, lipophilicity (logP, octanol-water) could be reasonably estimated using AP (r= 0.85) and ED (r= 0.85) methods for K= 5.  

Four topologically based methods were used by Basak and GrunwaW in estimating the boiling point of a set of 139 hydrocarbons and a group of 15 nitrosamines using the nearest neighbor K = 1). 

Basak and Grunwald also used topologically based similarity for KNN estimation of the mutagenicity of a set of 95 aromatic amines and the boiling point of a group of over 2900 chemicals with good results. 

Geometrical parameters, such as molecular shape parameters, sterimol descriptors, volume, bulk parameters, and 3D Wiener index, have been developed and used in SARs. Such parameters are derived from the relative distances of atoms in the 3D Euclidean space. We give below some examples of QSARs using 3D descriptors. 

---