Learning to taste: Application of deep learning to predict the sweetness of small organic molecules

Atif Afzal, Senior Scientist

With the recent developments in machine learning techniques, we use the data available on the taste of existing chemistries to develop efficient data-driven models for taste prediction. We demonstrate that these advanced machine learning models are highly efficient in classifying the molecules as bitter/sweet. Using the data and the model developed in this work, we can not only predict the sweetness of new molecules but also identify underlying relationships that distinguish the molecules as bitter/sweet.