A Chemistry-Informed Generative Deep Learning Approach for Enhancing Voltammetric Neurochemical Sensing in Living Mouse Brain

Abstract

Exploring the time-resolved dynamics of neurochemicals is essential for deciphering neuronal functions, intercellular communication, and neurophysiological or pathological mechanisms. However, the complex interplay among neurochemicals between neurocytes, coupled with extensive chemical signal crosstalk, puts simultaneous sensing of multiple neurochemicals into a longstanding challenge. Herein, we report a chemistry-informed generative neural network (CIGNN) model to separate the Faradaic and the non-Faradaic components from voltammetric currents, minimizing their mutual interference and enhancing quantitative accuracy. With the assistance of generative deep learning, we successfully establish a new platform for in vivo neurochemical sensing, which is validated by simultaneously monitoring the dynamics of dopamine (DA), ascorbic acid (AA), and ionic strength in a neuroinflammation mouse model. We observe that the stimulation with KCl solution triggers a significant enhancement of AA efflux on the model mice (300 ± 50 μM) compared with that from the control mice (170 ± 20 μM), as well as a significant decrease of ion influx (55 ± 7 mM) compared with that from the control mice (120 ± 16 mM), while not evoking a significant change in the DA release from the model mice (2.8 ± 0.3 μM) versus that from the control mice (3.0 ± 0.5 μM). This work provides a robust tool for studying multineurochemical signaling and elucidating the molecular mechanisms underlying various brain activities.