PharmaGNN: a model for odor prediction based on graph neural networks.

Area of Science:

• Computational chemistry
• Artificial intelligence
• Sensory science

Background:

• AI-assisted odor prediction is crucial for fragrance and food industries.
• Existing models struggle with limited features and imbalanced odor data.
• PharmaGNN integrates pharmacophore features and uses asymmetric loss to address these issues.

Purpose of the Study:

• To develop an accurate AI model for odor prediction.
• To improve molecular feature representation for olfaction.
• To mitigate class imbalance in odor datasets.

Main Methods:

• Developed PharmaGNN, integrating pharmacophore features.
• Employed an asymmetric loss function for imbalanced data.
• Evaluated model performance against existing methods.

Main Results:

• PharmaGNN achieved state-of-the-art performance with the highest AUROC score.
• The model demonstrated robust generalization across 232 odor labels.
• Pharmacophore features and molecular graph information significantly improved accuracy.

Conclusions:

• PharmaGNN accelerates novel flavor molecule design and enhances quality control.
• The model bridges cheminformatics and structural biology for rational scent design.
• PharmaGNN offers a tool for the digitization of olfaction.