Intense bitterness of molecules: machine learning for food and drugs

The bitter taste is one of the basic taste modalities and characteristic of many therapeutic drugs.1 While many health-beneficial food nutrients are known to be bitter,2 the intensely bitter compounds still constitute a problem in the pharmaceutical industry. The aversive taste of very bitter drugs can cause swallowing difficulties and compliance problems especially in pediatric medicine.3 Here we present "BitterIntense", a machine learning tool that classifies molecules into very bitter or not very bitter, based on their chemical structure.

The model was trained using chemically diverse compounds, reaching accuracy above 80% in predicting the bitterness intensity. BitterIntense suggested that intense bitterness does not indicate hepatotoxicity and application of BitterIntense across different datasets suggested that the abundance of very bitter compounds is more common among natural compounds of microbial and plant origin than in medicinal drugs and food. Screening of the human metabolome by BitterIntense has revealed potential ligands for extra-oral bitter taste receptors.

BitterIntense allows the prediction of bitterness levels of compounds as part of the drug development process in pharma industries and elucidation of ligands for extra-oral bitter taste receptors.

1. Levit, A. et al. The bitter pill: Clinical drugs that activate the human bitter taste receptor TAS2R14. FASEB J. 28, 1181–1197 (2014).

2. Drewnowski, A. & Gomez-Carneros, C. Bitter taste , phytonutrients , and the consumer : a review 1 – 3. Am. J. Clin. Nutr. 1424–1435 (2000). doi:10.2989/10220119.2012.694120

3. Mennella, J. A. & Beauchamp, G. K. Optimizing oral medications for children. Clin. Ther. 30, 2120–2132 (2008).