Decoding the quantitative structure-activity relationship and astringency formation mechanism of oxygenated aromatic...

Area of Science:

• Food Science
• Computational Chemistry
• Sensory Science

Background:

• Astringency, a key mouthfeel sensation, is challenging to evaluate conventionally due to cost and inefficiency.
• Quantitative Structure-Activity Relationship (QSAR) modeling offers a scalable computational alternative for predicting sensory properties based on molecular structure.

Purpose of the Study:

• To develop and validate a QSAR model for predicting astringency thresholds of oxygenated aromatic compounds.
• To identify key molecular descriptors influencing astringency.
• To discover novel natural astringent compounds using computational methods.

Main Methods:

• Collected 54 oxygenated aromatic compounds and performed molecular fingerprint similarity (MFS)-based clustering.
• Constructed six machine learning regression models, with AdaBoost showing the best performance (R²=0.778).
• Utilized Shapley Additive exPlanations (SHAP) for model interpretation, identified key descriptors (BCUT2D_LOGPLOW, VSA_Estate1), and employed Maximum Common Substructure (MCS) for compound identification.

Main Results:

• The AdaBoost model achieved high accuracy in predicting astringency thresholds, validated by sensory experiments.
• SHAP analysis identified critical molecular descriptors influencing astringency.
• Two natural astringent compounds were identified and their thresholds accurately predicted.
• Molecular dynamics (MD) simulations revealed hydrogen bonding and hydrophobic interactions drive protein-ligand aggregation.

Conclusions:

• QSAR modeling provides an efficient and accurate method for predicting astringency thresholds.
• Integration of QSAR and MD simulations facilitates the discovery and development of astringent compounds for the food industry.
• This approach advances predictive frameworks for astringency-focused food development.