Prioritizing Direct Photolysis Products Predicted by the Chemical Transformation Simulator: Relative Reasoning and Absolute Ranking

The United States Environmental Protection
Agency’s Chemical
Transformation Simulator (CTS) platform implemented the first freely
available reaction library to predict direct photolysis products of
organic contaminants in aquatic systems. However, the initial version
of the reaction library did not differentiate the formation likelihood
of each predicted product, and therefore, the number of predicted
products that are not observed tended to exponentially increase with
the prediction generation. To alleviate this problem, we first employed
relative reasoning algorithms to remove unlikely products. We then
ranked different reaction schemes according to their transformation
kinetics and removed slowly forming products. Applying the two strategies
improved the precision (the percentage of correctly predicted products
over all predicted products) by 34% and 53% for the internal evaluation
set and the external evaluation set, respectively, when products from
three generations were considered. This improved library also revealed
new research directions to improve predictions of the dominant phototransformation
products.