Developing a decision support system for chemical substitution using deep neural networks

To achieve a safe and sustainable use of chemicals it is necessary to identify and substitute hazardous and persistent chemicals.

This identification is often hindered by a lack of information, leading both to delayed decisions and to regrettable substitutions which do not lower the total risk.

We will construct a decision support system for chemical substitution, based on scores for both human and environmental risks as well as persistence.

To provide these scores we will construct a modelling pipeline which considers use/function, exposure, hazard and persistence. We will develop models for hazard and persistence based on artificial intelligence.

The modelling will be performed by a combination of chemical fingerprints and inputs on toxicity gathered e.g., from the US EPA, REACH registration dossiers and pesticide registration information.

To examine the linkages between low-level biochemical assays and high level individual/population measures we will also include data from biological assays from the US EPA ToxCast.

Finally, we will include product use and categories to ensure that suggested alternatives are not only dependent on chemical similarity.

The two main outcomes from this project is one database of compounds from international chemical inventories that due to their predicted risk are candidates for substitution and one open-source tool where users can input information on a compound and receive a short list of potential substitutes currently on the market.