Fast computation of marginalized walk kernel for virtual screening in drug discovery

Abstract

We introduce a new kernel function called Fast computation of marginalized walk kernels for graph classification for the virtual screening in drug discovery. This kernel is an extension of marginalized graph kernels. In this paper we use two dimensional graphical representations of chemical structures. The similarity between molecules is measured by using kernel function. For reducing the time complexity of kernel computation we use fast computing methods like Sylvester equation. The efficiency is evaluated using MUTAG dataset. The results show that the new kernel function is more efficient than existing kernel. This gives the same accuracy with lesser time complexity.