1 Walk-based Graph Kernel for Drug Discovery: A Review

The key motivation for the study of virtual screening is to reduce the time and cost requirement of the drug discovery process. Virtual screening is a computational method for finding an efficient drug molecule from pool of potential candidates. There are two different methods for virtual screening 1) structure based 2) ligand based. In the structure based method, 2D or 3D structure of a target molecule is used to screen the ligands which do not bind to the target molecule. Ligand based virtual screening is based on the fact that ligands similar to an active drug molecule might be active. The amount of information required is different in both the case. Structure based virtual screening is computationally intensive and complex while a few active ligand information is enough to start ligand based virtual screening. Based on the type of information, ligand based virtual screening can be performed in different ways. The machine learning approach using molecular graphs has been found to be very effective. Graph kernel is the similarity measure used to screen molecular graphs based on the structure. It is based on the fact that structurally similar molecules will have same property. In this review we have summarized the recent development in graph kernel for chemical molecule and elaborated upon the need of more accurate and efficient graph kernel with less computational complexity. The accuracy of different methods have been compared using standard dataset. The review shows the current state of art in the ongoing research in the design of efficient walk kernels