Neurotransmitter Concentration Prediction Based on Tensor-PLSR vs PLSR model: Comparison Study on Model Predictivity and Interpretability

Dopamine and serotonin concentrations in the brain extracellular environment play an important role in monitoring the signs of neuropsychiatric and neurological disorders, including major depressive and anxiety disorders, schizophrenia,and Parkinson’s disease. Many voltammetry methods have been developed to monitor neurotransmitter concentration in vivo. In order to accurately measure phasic changes in neurotransmitter levels which are rapid (milliseconds to seconds), fast scan cyclic voltammetry (FSCV) is developed to achieve sub-second measurement. However, FSCV suffers from the need for background subtraction. To keep the information in background and allow determination of the basal level of neurotransmitters, our group have previously developed a rapid pulse voltammetry (RPV) method.

Additionally, a machine learning model is coupled with voltammetry data to train the model and produce accurate predictions of analyte concentration. Principal components regression (PCR) has been used to reduce dimensionality in FSCV data and make predictions. Another dimensionality reduction method widely used in chemometrics is partial least squares regression (PLSR), which was shown to improve predictive accuracy over PCR when analyzing FSCV data. Tensor-PLSR is able to reduce multidimensional data. Traditionally, the experiment parameters are merged to make the data a matrix (2-mode tensor), sacrificing the experimental context reflected by the structure. However, the data could be interpreted as a 3-mode tensor (cube), and the comparison of prediction accuracy and interpretability between 3D-Tensor-PLSR and 2D-PLSR of RPV data is studied, and the preserved structure could reveal more insight of neurotransmitter redox reaction mechanism . The evaluation of model performance is not only based on prediction between dopamine concentrations, but also include its ability to discriminate between pH change and cation concentration in the buffer.