Neurochemical Concentration Prediction Using Deep Learning vs Principal Component Regression in Fast Scan Cyclic...

Area of Science:

• Neuroscience
• Analytical Chemistry
• Machine Learning

Background:

• Neurotransmitters like dopamine and serotonin regulate critical brain functions.
• Fast scan cyclic voltammetry (FSCV) is a key technique for in vivo neurochemical detection.
• Principal Component Regression (PCR) is commonly used for FSCV data analysis but has limitations in sensitivity and discrimination.

Purpose of the Study:

• To compare the performance of deep learning and PCR in identifying and quantifying neurotransmitter mixtures using FSCV data.
• To evaluate the accuracy of deep learning versus PCR for both in vitro and in vivo FSCV recordings.
• To assess the potential of deep learning in tracking neurochemical changes in the brain.

Main Methods:

• FSCV was used to record cyclic voltammograms of dopamine, epinephrine, norepinephrine, and serotonin in vitro.
• Data included single analytes and mixtures at various concentrations, as well as in vivo experiments.
• Deep learning models were developed and compared against conventional PCR for data analysis.

Main Results:

• Deep learning reduced identification accuracy errors by 5-20% compared to PCR for neurotransmitter mixtures.
• Both methods showed comparable performance for single analyte analysis.
• Deep learning demonstrated superior identification accuracy and discrimination performance for mixtures and in vivo data.

Conclusions:

• Deep learning offers a more reliable and accurate tool for analyzing FSCV data compared to PCR, especially for complex mixtures.
• The findings suggest deep learning can effectively track neurochemical changes, with potential applications in pharmacological studies.
• Further validation is recommended before widespread adoption of deep learning for FSCV analysis.