BitterSweet FAQs

BitterSweet provides the following services:
  • Categorical prediction of bitter and sweet tastes of small molecules.
  • Prediction of receptors linked to bitter-molecules.
  • Batch prediction of compounds (up to 200).
  • Structured exploration of curated information of bitter- and sweet-molecules.
  • Structured exploration of predicted information of bitter-sweet taste of molecules from Super Natural II, FooDB, FlavorDB, DrugBank and DSSTox.
BitterSweet has been tested to run successfully on the following systems:

OS Version Chrome Firefox Safari Microsoft Edge
Linux Ubuntu 16.04 71.0.3578.98 64.0 N/A N/A
MacOS Mojave 71.0.3578.98 64.0 12.0.2 N/A
Windows 10 71.0.3578.98 64.0 N/A 44.17763.1.0
While all the functionality of BitterSweet can be accessed using a mobile browser, BitterSweet is best viewed on mid to large screen sizes. 
We are using cookies to provide statistics that help us give best experience for our site.
BitterSweet: Building machine learning models for predicting the bitter and sweet taste of small molecules (2019)
Rudraksh Tuwani, Somin Wadhwa, Ganesh Bagler
Scientific Reports 9, Article number: 7155
doi: 10.1038/s41598-019-43664-y
Name Position Affiliation Contribution
Ganesh Bagler Project Head Center for Computational Biology, Indraprastha Institute of Information Technology (IIIT-Delhi), New Delhi Idea conception, Project design and management, Database design and implementation
Rudraksh Tuwani Research Assistant Center for Computational Biology, Indraprastha Institute of Information Technology Text Mining, Database design, Development of BitterSweet Web Resource, Data Visualisation and Data Analytics
Somin Wadhwa Research Intern Center for Computational Biology, Indraprastha Institute of Information Technology (IIIT-Delhi), New Delhi Predictive Modelling, Data Analytics, Development of BitterSweet Web Resource.
Front End: HTML, CSS, JavaScript, jQuery, Bootstrap, DataTables, Plotly, JSME Molecular Editor, and Jmol.
Back End: Python, Flask, Redis, MySQL, OpenBabel, RDKit, Scikit-learn, Pandas, NumPy.
You may contact us at for errata or any suggestions pertaining to the BitterSweet web server.
One or more chemical identifiers (SMILES / InChI / SDF / MOL File / PubChem ID / ZINC15 ID).
For information regarding interpretation of prediction output, please click here.
Due to the lack of compound-specific data mapping to sweet receptors which could form the basis of a predictive model, the BitterSweet server does not provide sweet receptor predictions.
Since we are using machine learning models in the backend to generate bitter-sweet predictions for molecules, it is to be noted that the applicability of these predictions is limited by the training data used for the models. For instance, if the database is limited to the molecules whose molecular weight is less than 1000, then the prediction for an unknown molecule whose molecular weight far exceeds that value might not be reliable.
To give an idea as to the applicability of predictions, we provide a visual overview (UMAP view) and a quantitative metric (Applicability Domain Statistic).
We use 0.5 as the threshold for categorizing a molecule as bitter/sweet.
We use 0.5 as the threshold for predicting whether a certain bitter-receptor is linked to a bitter compound or not.
Model Sensitivity Specificity F1-Score Area Under ROC Curve
Bitter/Non-Bitter 0.85 0.74 0.84 0.875
Sweet/Non-Sweet 0.641 0.96 0.771 0.852
Bitter Receptors 0.78 0.83 0.8 0.81
Search of molecules is available by the use of the following parameters -
  • Common Name
  • IUPAC Name
  • PubChem ID
  • Functional Group
  • Source of molecule: DrugBank, FooDB, FlavorDB, Super Natural II and DSSTox.
  • Bitter, Sweet, and Tasteless molecules
  • Linked Bitter Receptors
  • Structure Search (using JSME Molecular Editor)
  • Molecular Properties
The ‘functional group’ is an atom, or a group of atoms that has similar chemical properties whenever it occurs in different compounds. It defines the characteristic physical and chemical properties of families of organic compounds ( The functional groups were obtained using Checkmol, a free and an open source tool, checkmol, detect and assign the functional group information on any small molecules with 2D coordinates. The Checkmol is a command-line utility program, which reads molecular structure files in different formats and analyzes the input molecule for the presence of various functional groups. Analysis of Functional Groups in Organic Molecules, N. Haider, “Functionality pattern matching as an efficient complementary structure/reaction search tool: an open-source approach”, Molecules 15 (8) (2010) 5079–5092 ( List of functional groups generated by Checkmol can be viewed here.
The taste-information of all molecules present in BitterSweet can be downloaded from here.
The Tanimoto Coefficient of similarity between the fingerprints of query molecule and molecules present in BitterSweet is used as the criteria for finding similar molecules. We use Open Babel for calculating the molecular fingerprints.