Using molecular fingerprints and unsupervised learning algorithms to find simulants of chemical warfare agents

Abstract

The emergence of novel coronavirus highlights the importance of research and development of biological protective materials and functional protective equipment. As an important experimental material, the direct application of chemical warfare agents (CWAs) will cause great pollution to the environment. The effective search for simulants determines the process of CWAs experiments. This paper combines molecular fingerprint and unsupervised learning algorithm to develop a simulants selection framework. A selection strategy is developed based on the silhouette coefficient. The closest simulants are found (GA (TEP/DEEP), GB (DFP), GD (DEHP), HD (CEES), VX (Amiton)) under a threshold (Silhouette coefficient: 0.2). This study can effectively help researchers to find the best approximate simulant to a certain extent.