Discovery of chemical compound groups with common structures by a network analysis approach (affinity prediction method).

We developed a method in which the relationship between chemical compounds, characterized by the secondary dimensional descriptors by a standard method, is first determined by network inference, and then the inferred network is divided into the compound groups by network clustering. We applied this method to 279 active inhibitors of factor Xa found by the first screening. A large network of 266 active compounds connected with 408 edges emerged and was divided into 10 clusters. Surprisingly, the chemical structures that were common within the clusters, but diverse between them, could be extracted. The activity differences between the clusters provide rational clues for the systematic synthesis of derivatives in the lead optimization process, instead of empirical and intuitive inspections. Thus, our method for automatically grouping the chemical compounds by a network approach is useful to improve the efficiency of the drug discovery process.