Identification of Specific Substances in the FAIMS Spectra of Complex Mixtures Using Deep Learning

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Interpreting High-field Asymmetric Ion Mobility Spectrometry (FAIMS) spectra for single chemicals is straightforward.
• Identifying specific chemicals within complex mixtures using FAIMS presents significant challenges.

Purpose of the Study:

• To demonstrate the capability of a FAIMS system in detecting specific chemicals within complex mixtures.
• To develop and validate a deep learning model for enhanced chemical identification in mixtures.

Main Methods:

• A homemade FAIMS system was utilized to analyze pure substances (ethanol, ethyl acetate, acetone, etc.) and their mixtures.
• An EfficientNetV2 discriminant model was constructed and trained on the generated datasets.
• A blind test set was employed to rigorously verify the deep learning model's performance.

Main Results:

• The EfficientNetV2 model achieved convergence efficiently with a learning rate of 0.1 over 200 iterations.
• The trained model successfully identified specific substances in complex mixtures using the FAIMS system.
• Accuracies of 100% (ethanol), 96.7% (ethyl acetate), and 86.7% (acetone) were achieved in the blind test set, surpassing conventional methods.

Conclusions:

• Deep learning networks offer superior accuracy for FAIMS spectral analysis compared to traditional methods.
• The developed approach simplifies the FAIMS spectral analysis process.
• This research contributes to the advancement and broader application of FAIMS technology.