Many small-molecule medicines become toxic as a consequence of metabolism. Some of the most difficult to predict toxicities are caused by reactive metabolites, which covalently bind proteins. Alert structures are substructures that are prone to bioactivation, and the presence of alert structures is often used to flag which molecules might be bioactivated into reactive species. Compared to alert structures, we found that machine learning can better identify molecules that are prone to bioactivation, by modeling more precisely the nuances of metabolism, metabolite structures, and the subsequent reactivity of metabolites. More recently, we found that machine learning can also identify which substructures are prone to bioactivation, thereby identifying novel and emerging structural alerts of increased importance in recent years. There still remain key questions about how to best make use of machine learning in drug development, but the potential for substantial impact is high.

Using machine learning to understand structural alerts and the bioactivation of medicines

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.