Managing Uncertainty on the Integration of Safety and Health Indexes in Computer-Aided Molecular Design

Abstract

In this work, a novel molecular design methodology has been developed to incorporate the safety and health aspects of potential molecules, while managing the uncertainty in property prediction models applied to measure the safety and health performance. In recent years, various computer-aided molecular design techniques have been developed to design molecules that meet customer requirements. To ensure the selected molecule does not bring harm to the consumers, the concept of inherent safety and health must be incorporated during the synthesis of the molecule. Several safety and health parameters are employed to examine the adverse safety and health impacts that may be posed by the synthesized molecule. These parameters are assessed by the physicochemical properties, which are estimated using property prediction models. Subindex scores are then assigned based on the properties to signify the hazard level. However, there exists uncertainty in the prediction models that can affect the accuracy of the allocated scores. The methodology developed in this work can manage the uncertainty on properties used in the safety and health parameters. The scores at the uncertain region are enhanced to account for uncertainty originated from the prediction models. The quantification of uncertainty ensures that the enhanced scores can better measure the inherent hazard level of the molecule. Hence, the molecular design techniques will be able to identify molecules with less hazards to use in a process without compromising the rest of the target properties.