Bitterness-Classifier

Bitterness Classifier – Predicting Taste from Chemical Structure

Abstract

The sense of taste (also known as gustation) is among the five traditionally recognized senses. This crucial chemical sense mediates between the external chemical and internal biochemical environments. Its prominent role comes in determining whether to accept incoming food to be swallowed or to reject it. Particularly, the bitter taste is believed to be a sign of the presence of poisonous compounds. Determining the presence of a bitter taste is important for Pharmaceutical companies especially when it comes to producing orally consumed pediatric therapeutics. In addition, the receptors for bitter taste are found in extra-oral tissues, including gastrointestinal tract and respiratory tract, and recently it was found that they represent unprecedented drug targets for asthma. Therefore, identifying bitterness has therapeutic potential as well. Bitter taste is complex: the estimations are that there are thousands or more of bitter compounds, and it is known that bitter compounds are very diverse in terms of their chemical structure. To date, there are not many CS-based solutions for predicting bitterness. Our project aims to benefit from the Machine Learning field in order to implement a classifier that is able to predict the bitter/non-bitter taste of chemical molecules. To this end we used different learning algorithms, such as Random Forests, and Neural Networks, using ML libraries (such as Scikit-learn) and data from Prof Niv lab: BitterDB, a large public repository of bitter compounds as positive set, and in house non-bitter set as negative set. We also developed a public site, which uses our model and predicts whether a given compound (represented by a set of descriptors) is bitter or not.
The outcomes will be useful for the both food and pharma industries.