A QSTR-Based Expert System to Predict Sweetness of Molecules

Area of Science:

• Computational chemistry
• Cheminformatics
• Sensory science

Background:

• Predicting chemical properties like taste from molecular structure is crucial for drug discovery and food science.
• Existing Quantitative Structure-Activity Relationship (QSAR) models require rigorous validation, adhering to established guidelines.

Purpose of the Study:

• To develop a novel Quantitative Structure-Taste Relationship (QSTR) model for predicting the sweetness of chemical compounds.
• To create a validated expert system for assessing molecular taste based on advanced molecular similarity.
• To enhance the understanding of the relationship between chemical structures and their perceived sweetness.

Main Methods:

• Utilized advanced molecular similarity techniques, specifically conformation-independent extended-connectivity fingerprints (ECFPs), to analyze 649 sweet and non-sweet molecules.
• Developed a Quantitative Structure-Taste Relationship (QSTR) model adhering to the five OECD principles for (Q)SAR validation.
• Employed a consensus approach combining Partial Least Squares-Discriminant Analysis and N-nearest-neighbor classification for complex taste assignments.
• Validated the model using Monte Carlo procedures and an external dataset.

Main Results:

• Molecular similarity in the ECFP space demonstrated a strong correlation with molecular taste.
• The developed QSTR expert system achieved satisfactory predictive performance, comparable to state-of-the-art models.
• The model effectively handled molecules in subspaces where taste assignation was challenging.

Conclusions:

• The novel QSTR model provides a robust method for predicting chemical sweetness based on molecular structure.
• The system offers valuable insights into structure-taste relationships, facilitating the rational design of novel sweeteners.
• The adherence to OECD principles ensures the reliability and applicability of the developed QSTR model.