Towards quantifying the uncertainty in in silico predictions using Bayesian learning

•Predicting quantitative values and uncertainties are important for computational toxicology algorithms in risk assessment.•Bayesian learning neural networks have been trained to provide these predictions for human molecular initiating events.•These uncertainties are shown to be able to distinguish between training set, similar and decoy compounds.•The ability of in silico algorithms to make predictions including uncertainty is key to next generation risk assessment.