Is bitter actually better? machine learning predictions and analysis of tastants

Eitan Margulis ¹ Yuli Slavutsky ² Anne Tromelin ³ Yuval Benjamini ² Masha Y. Niv ¹

¹The Institute of Biochemistry, Food Science and Nutrition, Robert H. Smith Faculty of Agriculture, Food and Environment, The Hebrew University of Jerusalem, Rehovot, Israel
²Center of Interdisciplinary Data Science Research, The Hebrew University of Jerusalem, Jerusalem, Israel
³Centre Des Sciences Du Goût Et De L'alimentation, Agrosup Dijon, CNRS, INRA, Université Bourgogne Franche-Comté, F-21000, Dijon, France

Numerous compounds elicit bitter taste sensation via activation of G-protein coupled receptors from the T2R sub-family. Due to their chemical diversity, as evident in the BitterDB database¹, bitter compounds may have additional biological targets, while T2Rs may have ligands and physiological roles beyond taste. This raises questions regarding the multi-functionality of both T2Rs and their ligands. We apply machine learning tool BitterPredict² for predicting the identity and abundance of bitter compounds in several datasets of compounds with a specific biological activity. Our results indicate that bitterness can be correlated with other physiological phenomena including reduced risk for hepatotoxicity, distinctive smell and more. Since there are 25 subtypes of T2Rs in human, the possibility of distinctive roles for different subtypes is of interest. We present a ligand – receptor recommendation system, BitterMatch, that assigns bitter compounds to individual T2Rs based on molecular properties and on previously known ligand-receptor associations.

Reference:

Dagan-Wiener, A. et al. BitterDB: taste ligands and receptors database in 2019. Nucleic Acids Res. 1–7 (2018). doi:10.1093/nar/gky974
Dagan-Wiener, A. et al. Bitter or not? BitterPredict, a tool for predicting taste from chemical structure. Sci. Rep. 7, 1–13 (2017).