Characterizing bitterness: identification of key structural features and development of a classification model.

Area of Science:

• Computational Chemistry
• Cheminformatics
• Structure-Activity Relationship Studies

Background:

• Understanding the molecular basis of taste, particularly bitterness, is crucial for drug design and food science.
• Existing methods for predicting bitterness are limited in scope and accuracy.
• A comprehensive classification system for bitterness in small molecules is needed.

Purpose of the Study:

• To develop and validate a computational method for classifying bitter small molecules.
• To identify specific substructural features correlated with bitterness.
• To build a predictive model for bitterness based on these features.

Main Methods:

• Utilized a dataset of 649 bitter and 13,530 random molecules from the MDL Drug Data Repository (MDDR).
• Employed circular fingerprints (MOLPRINT 2D) and information-gain feature selection to identify key substructures.
• Applied a Naïve Bayes classifier for bitterness prediction.

Main Results:

• Identified statistically significant substructural features associated with bitterness, including sugar moieties and branched carbon scaffolds.
• Classified bitter compounds into two distinct groups: larger, oxygen/carbon-rich molecules (often with sugar moieties) and smaller molecules with nitrogen/sulfur fragments.
• Achieved 72.1% prediction accuracy for bitter compounds, with feature selection improving false-positive reduction.
• The model correctly identified some bitter compounds (e.g., cynaropicrine) while not misclassifying others (e.g., promethazine, saccharin).

Conclusions:

• The developed method provides a reliable approach for classifying bitterness in small molecules.
• The identified substructural features are functionally responsible for the bitter taste.
• This work presents a large database of bitter compounds and a validated predictive model.