Harnessing Explainable AI to Explore Structure-Activity Relationships in Artificial Olfaction.

Area of Science:

• Artificial olfaction
• Chemical sensing
• Materials science

Background:

• Chemical sensor arrays aim to replicate mammalian olfactory systems for artificial olfaction.
• Developing effective receptor materials requires understanding structure-activity relationships.

Purpose of the Study:

• To visualize the relationship between sensing signal features and odorant molecular features using explainable AI (XAI).
• To establish guidelines for developing effective receptor materials for artificial olfaction.

Main Methods:

• Utilized explainable AI (XAI) techniques, specifically convolutional neural networks (CNN) and Score-Class Activation Mapping (Score-CAM).
• Analyzed 94 odor samples prepared from pure solvents for classification tasks.

Main Results:

• Successfully visualized the structure-activity relationship between odorant molecules and sensor signals.
• Accurately extracted information on active receptor materials and data points within sensor signals.
• Demonstrated the effectiveness of XAI in analyzing sensor signals for odor data.

Conclusions:

• XAI techniques are crucial for advancing artificial olfaction by enabling detailed analysis of sensor signals.
• Visualizing structure-activity relationships aids in the rational design of novel olfactory receptor materials.