VirtualTaste

VirtualTaste predicts the taste profiles (sweet, bitter, and sour) of chemical compounds to assess organoleptic properties for food and pharmaceutical research.

Key Features:

• Taste Prediction Models: Employs machine learning models to predict three taste endpoints: sweet, bitter, and sour.
• Performance and Validation: Reported overall accuracy of 90% and Area Under the Curve (AUC) of 0.98 in 10-fold cross-validation and on independent test sets.
• Molecular Representations and Outputs: Uses molecular fingerprints and returns per-compound confidence scores.
• Comparative Analysis: Cross-references predictions with similar compounds from the training set to provide known-activity comparisons.
• Bitter Receptor Target Prediction: Provides target-prediction insights for predicted bitter compounds against 25 known bitter receptors.

Scientific Applications:

• Pharmaceutical Industry: Identifies potential bitterness of drug candidates to help anticipate and mitigate patient adherence issues.
• Food Industry: Evaluates and supports optimization of sweet, bitter, and sour profiles during food product development.

Methodology:

Applies machine learning on molecular fingerprints, cross-referencing a training set of compounds with known tastes; performance was assessed by 10-fold cross-validation and independent test sets; target prediction was used to evaluate interactions with 25 bitter receptors.