Porträttbild på Per-Olof Syrén. Bild från SciLifeLab
Better models are needed to predict the environmental toxicity of chemicals
Digital tools can help identify potentially hazardous chemicals early in the development process. A new study within Mistra SafeChem and the European partnership program PARC shows, however, that current models have limitations when applied to certain groups of chemicals.
The researchers used computational tools to predict the ecotoxicity of around 19,000 molecules, including acids, amines and amides. These chemical groups are widely relevant in chemical production, for example in pharmaceuticals and polymers.
Almost 99.9 percent of the molecules were predicted as non-toxic. However, the reliability of most predictions was low to moderate. One important reason was that many of the chemicals fell outside the chemical space represented by the data used to train currently available models. The result should therefore be interpreted as a screening result with limited reliability, rather than evidence that the substances are non-toxic.
– Our study highlights an important limitation: computational screening has great potential for SSbD, but the models need to be better adapted to the chemical domain being assessed for the results to be used with confidence, says Per-Olof Syrén, Professor at KTH and leader of Case study 5 in Mistra SafeChem.
The study highlights the need for new and improved models tailored to these abundant chemical groups. To achieve this, new experimental data of high quality is required that together could improve predictive capacity of models and strengthen the use of digital tools within SSbD, where safety and sustainability are considered early in the development of new chemicals and materials.
Read the scientific article:
Applying in silico ecotoxicity prediction to extensive datasets to support safer and more sustainable discovery chemistry External link, opens in new window.