Novel Hierarchical Classification and Visualization Method for Multiobjective Optimization of Drug Properties: Application to Structure-Activity Relationship Analysis of Cytochrome P450 Metabolism

Publication: J Chem Inf Model
Software: ADMET Predictor®


In the lead optimization process, medicinal chemists must consider various chemical properties of active compounds, including ADME/Tox properties, and find the best compromise among these. This study presents a novel data mining method for multiobjective optimization of chemical properties, which consists of the hierarchical classification and visualization of multidimensional data. A hierarchical classification tree model is generated by an extension of recursive partitioning that utilizes averaged information gains for multiple objective variables as a quality-of-split criterion. All the hierarchically structured data objects are represented using a large-scale data visualization technique. The technique is an extension of HeiankyoView, which displays data objects as colored icons and group nodes as rectangular borders. Each icon is divided into subregions with different colors, so that it can present multidimensional data according to brightness of the colors. The proposed method was applied to the structure−activity relationship analysis for cytochrome P450 (CYP) substrates. The substrate specificity of six CYP isoforms was successfully delineated:  e.g., CYP2C9 substrates are anionic compounds, while CYP2D6 substrates are cationic; and CYP2E1 substrates are smaller compounds, while CYP3A4 substrates are larger compounds.