Classification of Odor-Derived Electroantennograms with Machine Learning.

Area of Science:

• Insect olfaction research
• Biosensor development
• Machine learning applications

Background:

• Insects detect volatile organic compounds (VOCs) using olfactory receptors on their antennae.
• Antennal local field potential (LFP) responses to VOCs have traditionally been used for concentration measurement.
• Recent advancements suggest LFPs may also be used for VOC discrimination and identification.

Purpose of the Study:

• To investigate the potential of using antennal LFP time-series responses for VOC classification.
• To capture key LFP characteristics like waveform dynamics, intensity, slope, and duration for analysis.
• To demonstrate the feasibility of using machine learning for VOC identification from antennal responses.

Main Methods:

• Recording LFPs from excised *Manduca sexta* moth antennae exposed to floral and disease-associated VOCs.
• Extracting principal components from LFP time-series data to represent response characteristics.
• Training machine learning models (support vector machines, random forests) on LFP data for classification.

Main Results:

• Machine learning models successfully predicted and classified individual VOCs across various concentrations.
• The models could also classify complex VOC mixtures based on their elicited LFP waveforms.
• Antennal olfactory responses were shown to be effective for classifying VOC concentration, identity, and duration.

Conclusions:

• Antennal LFPs contain rich information for classifying VOCs, extending beyond simple concentration detection.
• This approach has significant implications for developing advanced chemical sensing technologies.
• Potential applications include environmental monitoring, agricultural pest detection, and disease diagnostics.