Applying in silico ecotoxicity prediction to extensive datasets to support safer and more sustainable discovery chemistry.

Abstract

Amide bonds are highly present in many chemical products, including pharmaceuticals and polymers. They are constituted from amines and acids, both of which are abundant in the chemical industry. Therefore, there is a need to bring these chemicals into alignment with the recent European Commission (EC) Safe and Sustainable by Design (SSbD) framework, and its practices. This study investigates the ecotoxicity potential of an extensive dataset of approximately 19,000 molecules, including amines, acids and amides, using Quantitative Structure Activity Relationship (QSAR) models, VEGA and JANUS, respectively. Most predictions showed low to moderate reliability, indicating an outside-the-applicability domain pattern of the analyzed compounds. Under these conditions, 99.9% of the dataset was classified as non-toxic, which should be interpreted as a limited-reliability screening result. This pattern was further explored by analyzing the chemical space of both marketed chemicals and the training sets of the models used. Overall, this analysis highlights the need to develop new QSAR models tailored to amines and acids, to overcome the out-of-domain challenges and support the operationalization of the SSbD framework.