Preparation of Simplified Molecular Input Line Entry System Notation Datasets for use in Convolutional Neural Networks

Abstract

Simplified Molecular Input Line Entry System (SMILES) is a type of chemical notation. The SMILES format allows the representation of chemical structures in a shape easily readable by computer programs. This allows many techniques, such as Artificial Neural Networks (ANNs) to be applied on the SMILES formatted data. One of the highest-performing ANN types is the Convolutional Neural Networks (CNNs), designed to work on images or matrix-shaped data. In this paper, the authors will present the preparation of the SMILES dataset for use by CNNs. The paper will start with a brief description of the SMILES format, followed by the explanation of the dataset transformation into an NPY matrix-based format, with an example of utilization via the application of popular CNN architectures on a transformed dataset. The proposed architecture achieves satisfactory results (AUC=0.92), with the transformation algorithm speed also proving satisfactory (0.08 seconds per data point)