My research develops tools to enable understanding of how information is encoded in chemical messengers (neurotransmitters; e.g., serotonin, dopamine) in the brain by measuring them in real-time, in awake, behaving mice models. A molecular scale understanding of complex behavior and brain (dys)function would enable better treatments for neuropsychiatric disorders such as anxiety and depression (which affect 50% of families worldwide). A major challenge is the lack of analytical techniques that can detect multiple neurotransmitters simultaneously, across physiologically and behaviorally relevant time scales (ms, min, h). To date, the most suitable technique is voltammetry, which applies a specific voltage waveform to a micron-sized brain-implantable sensor. However, voltammetry is limited by the number of analytes and timescales that can be achieved for in vivo monitoring, impeding an understanding of how our brains encode information across neurotransmitters in a coordinated manner.
My graduate work involves developing a new voltammetry technique that combines novel voltammetry waveforms with machine learning. The technique is called rapid pulse voltammetry with partial least squares regression (RPV-PLSR), which utilizes a new type of pulse waveform and data analysis approach. Using RPV-PLSR, we published a proof-of-concept study to monitor serotonin and dopamine at both basal and stimulated levels, simultaneously, and compared these results to conventional approaches. We showed the utility of faradaic and non-faradaic current responses to build robust statistical models. The unique electrochemical signals present in pulse, rather than sweep, waveforms provide valuable chemical information in the background current that other methods do not generate.
However, the design of RPV waveforms (and voltammetry waveforms in general) is an arduous and inefficient process. To address this challenge, I developed a machine learning-guided approach for systematic design of novel-yet-optimal voltammetry waveforms using Bayesian optimization. Here, I identified optimal serotonin detection RPV waveforms under various physiologically relevant conditions.
Lastly, the custom waveforms and data analysis procedures used for RPV required the development of several open-source software solutions for user-friendly acquisition, analysis, storage, and sharing of voltammetry data at scale, which are reported herein. I outline how future extensions of fast voltammetry and machine learning can address domain generalizability (i.e., the ‘beaker-to-brain’ issue), and how the techniques developed herein will broaden the scope of multi-analyte electroanalytical method development.
