Development of Adverse Outcome Pathway for PPARγ Antagonism Leading to Pulmonary Fibrosis and Chemical Selection for Its Validation: ToxCast Database and a Deep Learning Artificial Neural Network Model-Based Approach.

Abstract

Exposure to certain chemicals such as disinfectants through inhalation is suspected to be involved in the development of pulmonary fibrosis, a lung disease in which lung tissue becomes damaged and scarred. Pulmonary fibrosis is known to be regulated by transforming growth factor β (TGF-β) and peroxisome proliferator-activated receptor gamma (PPARγ). Here, we developed an adverse outcome pathway (AOP) to better define the linkage of PPARγ antagonism to the adverse outcome of pulmonary fibrosis. We then conducted a systematic analysis to identify potential chemicals involved in this AOP, using the ToxCast database and deep learning artificial neural network models. We identified chemicals bearing a potential inhalation hazard and exposure hazards from the database that could be related to this AOP. For chemicals that were not present in the ToxCast database, multilayer perceptron models were developed based on the ToxCast assays related to the AOP. The reactivity of ToxCast untested chemicals was then predicted using these deep learning models. Both approaches identified a set of chemicals that could be used to validate the AOP. This study suggests that chemicals categorized using an existing database such as ToxCast can be used to validate an AOP and that deep learning approaches can be used to characterize a range of potential active chemicals for an AOP of interest.