Chapter 8: Virtual Screening with Convolutional Neural Networks

In this chapter, we examine how convolutional neural networks (CNN), a type of deep learning methodology, can be applied to the evaluation of protein–ligand complexes in silico. CNNs have been the primary driver of the enormous progress in computational analysis of digital images and videos (computer vision) over the last few years. This success has sparked interest in, and the use of, CNNs across a diverse range of application areas. Recently there have been several applications of CNNs to the scoring of protein–ligand complexes, in particular to assess whether a ligand will bind (virtual screening), its binding mode (pose prediction), and the strength of binding (binding affinity prediction). The focus of this chapter is on the first of these, virtual screening, but the techniques and problem formulation are transferable to other tasks.