QuantumScents: Quantum-Mechanical Properties for 3.5k Olfactory Molecules.

Area of Science:

• Computational chemistry
• Cheminformatics
• Olfactory research

Background:

• Quantitative structure-odor relationships (QSORs) are vital for understanding olfaction.
• Existing datasets lack essential molecular feature data for QSOR studies.
• The Leffingwell dataset provides expert-labeled odorant molecules but requires computational features.

Purpose of the Study:

• To introduce QuantumScents, a novel dataset augmenting the Leffingwell dataset with quantum mechanical features.
• To provide a comprehensive resource for QSOR research, including molecular coordinates and electronic properties.
• To enable the development of predictive models for molecular odor classification.

Main Methods:

• Augmented the Leffingwell dataset with quantum mechanics calculations (PBE0 functional).
• Generated 3D coordinates, total energy, dipole moments, and per-atom Hirshfeld charges, dipoles, and ratios for over 3.5k molecules.
• Trained a Message Passing Neural Network (MPNN) using chemprop for molecular classification based on scent labels.

Main Results:

• The QuantumScents dataset comprises 3.5k diverse molecules with detailed quantum mechanical properties.
• Hirshfeld charges and ratios were found to contain sufficient information for accurate molecular odor classification.
• A Message Passing Neural Network trained on QuantumScents demonstrated predictive capabilities for scent labels.

Conclusions:

• QuantumScents provides a valuable resource for advancing QSOR research.
• Atomic-level electronic properties derived from quantum mechanics are effective predictors of molecular odor.
• The developed methodology and dataset facilitate the creation of sophisticated olfaction models.