A breakthrough in identifying weak areas of chemical prediction models

Scientists at the University of Vienna have created a software tool, ‘MolCompass’, to identify the limits of machine learning models used in chemical risk assessment. The hope is to boost trust in computational methods among toxicologists and regulators. Born out of the EU-supported RISK-HUNT3R project, the tool flags up model failures, allowing toxicologists to explore ‘chemical space’ visually. The findings have been published in the Journal of Cheminformatics.