Case Study: IFF improves sensory property prediction with Intellegens Alchemite™ deep learning

IFF - Customer Case Study

IFF needed a better way to predict sensory properties of chemical compounds, a difficult and expensive task because the measurements depend on human panels and the available physicochemical and sensory data were sparse. Intellegens worked with IFF and Optibrium using its Alchemite™ deep learning approach to tackle this challenge.

Intellegens applied Alchemite™ to impute sparse data and predict sensory properties, outperforming conventional QSAR methods and an alternative graph convolutional neural network. The model delivered higher predictive accuracy, with R² improvements of 0.26 to 0.45 over the next best method, while also providing reliable uncertainty estimates and better identification of activity cliffs, helping IFF select compounds more effectively and save experimental time and resources.

IFF

Dmitriy Chekmarev

Senior Research Investigator