PredCoffee: A binary classification approach specifically for coffee odor.

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in chemistry

Background:

• Traditional methods for identifying odorant molecules are often costly and time-consuming.
• Machine learning offers a promising alternative for efficient molecular property prediction.
• Predicting specific sensory attributes like coffee odor requires specialized models.

Purpose of the Study:

• To develop a machine learning model for predicting coffee odor in molecules.
• To identify underlying regularities in molecules possessing a coffee aroma.
• To create a cost-effective and accurate binary classifier for coffee odor detection.

Main Methods:

• Collection of a dataset including 371 coffee-odor molecules and 9,700 non-coffee-odor molecules.
• Training and evaluation of various machine learning models: Knowledge-guided Pre-training of Graph Transformer (KPGT), Support Vector Machine (SVM), Random Forest (RF), Multi-layer Perceptron (MLP), and Message-Passing Neural Networks (MPNN).
• Selection of the best-performing model for the final predictor.

Main Results:

• The Knowledge-guided Pre-training of Graph Transformer (KPGT) model demonstrated superior performance.
• The KPGT model achieved a prediction accuracy exceeding 0.84.
• The developed predictor was successfully deployed as a webserver named PredCoffee.

Conclusions:

• Machine learning, particularly KPGT, provides an effective and accurate method for predicting coffee odor in molecules.
• The PredCoffee webserver offers a valuable tool for researchers and industry professionals.
• This approach can significantly reduce costs and time associated with traditional odor assessment methods.