Molecular Similarity Based on Functional Groups: Subtree Kernels for Virtual Screening in Drug Discovery

Abstract

Current developments in computer-aided drug design (CADD) reduces the time and cost of drug discovery process. Using various computational techniques, large number of compounds in the chemical database are analyzed and screened. Virtual screening can be classified into structure based and ligand based methods. The amount of information required for structure based virtual screening is too high when compared to ligand based method. In ligand based methods, classification of active and inactive drugs can be done using Hidden Markov Model, Support Vector Machine (SVM), Clustering etc. SVM is widely used nowadays. Classification using SVM and graph kernel function is introduced recently and shows good accuracy over the other existing methods. Different graph kernel functions are used for finding similarity measure in SVM. Here, we propose a new method for finding the similarity based on the functional groups present in the molecules. The accuracy is compared with that of existing graph kernel methods using standard data sets (PTC and MUTAG). , Coimbatore, India

 

Keywords: drug discovery; virtual screening; SVM; molecular graph similarity functional group; tree kernel.