Integrating Experiments and Machine Learning Models: Examples from the ATOM Consortium

Predictive data-driven machine learning models are an important component of emerging approaches to accelerate molecular design. An integrated process for connecting targeted experiments with model algorithms and architectures is central to predictive performance and domain of applicability. The presentation will focus on two examples of this experiment-AI integration: 1) development of new drug-induced liver injury (DILI) models for assessing the safety of a proposed molecule; and 2) the development and use of efficacy, safety, and pharmacokinetic models for multiparameter design optimization of novel and selective histamine inhibitors.