A principal odor map unifies diverse tasks in olfactory perception.

Area of Science:

• Computational chemistry
• Neuroscience
• Machine learning

Background:

• Predicting odor perception from molecular structure is a significant challenge in olfaction research.
• Understanding structure-odor relationships is crucial for various applications, including fragrance development and safety assessments.

Purpose of the Study:

• To develop a predictive model for odor quality based on molecular structure.
• To create a principal odor map (POM) that captures perceptual relationships between odorants.
• To evaluate the model's performance against human panelists and existing chemoinformatic models.

Main Methods:

• Utilized graph neural networks (GNNs) to construct a principal odor map (POM).
• Trained the GNN model on a dataset of odorants and their associated odor qualities.
• Validated the model prospectively on 400 novel odorants, comparing its predictions to human panel data.

Main Results:

• The POM demonstrated high reliability in predicting odor quality, comparable to human perception.
• Model-generated odor profiles more accurately matched the mean of a trained human panel than individual panelists.
• The POM outperformed traditional chemoinformatic models in various odor prediction tasks.

Conclusions:

• The developed principal odor map (POM) successfully encodes generalized structure-odor relationships.
• This GNN-based approach enables accurate odor prediction for uncharacterized odorants.
• The study paves the way for the digital representation and prediction of odors.