Prediction of sweetness by multilinear regression analysis and support vector machine

Area of Science:

• Computational Chemistry
• Food Science
• Cheminformatics

Background:

• Sweetness is a key characteristic for food additives.
• Predicting sweetness quantitatively is crucial for industry applications.
• Understanding structure-sweetness relationships can guide new sweetener development.

Purpose of the Study:

• To develop and validate quantitative models for predicting the logarithm of sweetness (logSw).
• To identify molecular descriptors that influence sweetness.
• To assess the performance of multilinear regression (MLR) and support vector machine (SVM) for sweetness prediction.

Main Methods:

• A dataset of 320 compounds with known sweetness values was compiled.
• Compounds were characterized by 12 molecular descriptors.
• The dataset was split into training (214 compounds) and testing (106 compounds) sets.
• MLR and SVM models were employed to predict logSw.

Main Results:

• Both MLR and SVM models achieved high predictive accuracy on the test set.
• Correlation coefficients of 0.87 (MLR) and 0.88 (SVM) were obtained.
• Identified molecular descriptors offer structural insights into sweetness perception.

Conclusions:

• Quantitative structure-activity relationship (QSAR) models can effectively predict compound sweetness.
• The findings support the established AH/B System model of taste perception.
• These models can aid in the design and selection of novel sweetening agents.