Bitter or not? BitterPredict, a tool for predicting taste from chemical structure

Abstract

Bitter taste is an innately aversive taste modality that is considered to protect animals from consuming toxic compounds. Yet, bitterness is not always noxious and some bitter compounds have beneficial effects on health. Hundreds of bitter compounds were reported (and are accessible via the BitterDB http://bitterdb.agri.huji.ac.il/dbbitter.php), but numerous additional bitter molecules are still unknown. The dramatic chemical diversity of bitterants makes bitterness prediction a difficult task. Here we present a machine learning classifier, BitterPredict, which predicts whether a compound is bitter or not, based on its chemical structure. BitterDB was used as the positive set, and non-bitter molecules were gathered from literature to create the negative set. Adaptive Boosting (AdaBoost), based on decision trees machine-learning algorithm was applied to molecules that were represented using physicochemical and ADME/Tox descriptors. BitterPredict correctly classifies over 80% of the compounds in the hold-out test set, and 70–90% of the compounds in three independent external sets and in sensory test validation, providing a quick and reliable tool for classifying large sets of compounds into bitter and non-bitter groups. BitterPredict suggests that about 40% of random molecules, and a large portion (66%) of clinical and experimental drugs, and of natural products (77%) are bitter.

The taste of toxicity: A quantitative analysis of bitter and toxic molecules

“…Building upon the database of bitter compounds, the BitterDB , and the newly developed bitterness prediction tool BitterPredict , we attempt to quantify the bitterness‐toxicity relationship by analyzing the chemical spaces of known bitter and toxic compounds. Specifically, we ask “Are toxic compounds typically bitter?…”

BitterSweet: Building machine learning models for predicting the bitter and sweet taste of small molecules

“…However, inconsistencies in the curation process due to the inclusion of molecules with unverified taste information and incomplete representation of chemical space can lead to incorrect inferences and predictions. BitterPredict 19 and BitterX 13 have used unverified non-bitter molecules as a significant part of their training sets (55.6% and 50% respectively), which potentially adds noise to their models. While e-Bitter 21 , Rojas et al 16 and BitterSweetForest 22 mitigated this problem by utilizing only experimentally verified data, this significantly reduced the size of their datasets and might have led to insufficient representation of the bitter-sweet chemical space.…”

“…Tasteless compounds were included as important controls for both bitter and sweet taste prediction. The datasets were split into training and testing sets such that the latter corresponded to the external validation/test sets established by BitterPredict 19 for bitter/non-bitter prediction and Rojas et al 16 for sweet/non-sweet prediction ( Supplementary Table S1). The curated training dataset is structurally diverse when seen in comparison to random bioactive molecules from ChEBI 24 , as evident in the 2D t-SNE plot generated using the physicochemical features ( Figure 1), with molecules from different sources incrementally capturing subsets of the general chemical space.…”

“…Rojas et al 16 instead tackled the problem of sweet-taste prediction using experimentally verified data from the literature, 2D molecular descriptors and ECFPs (Extended Connectivity Fingerprints) from Dragon 17 towards the development of a QSTR-based expert system 16,18 . BitterPredict 19 enhanced the training data of BitterX 13 , by including molecules curated by Rojas et al 18 and those identified as being 'probably non-bitter' from Fenaroli's Handbook of Flavor Ingredients 20 , while excluding the random molecules. They further established rigorous external validation sets (curated from literature and other unpublished resources), to show the efficacy of their AdaBoost model based on physicochemical and ADMET properties from Canvas.…”

Bitterless guaifenesin prodrugs—design, synthesis, characterization, in vitro kinetics, and bitterness studies

“…[1,2] The most commonly used bitter-tasting drugs include paracetamol, ampicillin, azithromycin, diphenhydramine, erythromycin, ibuprofen, penicillin, pseudoephedrine, and guaifenesin. [3][4][5] Nowadays, the medicinal industry has recognized the importance of taste masking; however, the development of an appropriate formulation is relatively expensive and time-consuming. [6,7] A vast number of techniques have been invoked for concealing the unpleasant taste of drugs, [8][9][10] including polymer coating, complex formation with β-cyclodextrin, ion exchange resins, and solubility-reducing methods.…”