Identifying Neural Signatures of Dopamine Signaling with Machine Learning.

The emergence of new tools to image neurotransmitters, neuromodulators, and neuropeptides has transformed our understanding of the role of neurochemistry in brain development and cognition, yet analysis of this new dimension of neurobiological information remains challenging. Here, we image dopamine modulation in striatal brain tissue slices with near-infrared catecholamine nanosensors (nIRCat) and implement machine learning to determine which features of dopamine modulation are unique to changes in stimulation strength, and to different neuroanatomical regions. We trained a support vector machine and a random forest classifier to decide whether the recordings were made from the dorsolateral striatum (DLS) versus the dorsomedial striatum (DMS) and find that machine learning is able to accurately distinguish dopamine release that occurs in DLS from that occurring in DMS in a manner unachievable with canonical statistical analysis. Furthermore, our analysis determines that dopamine modulatory signals including the number of unique dopamine release sites and peak dopamine released per stimulation event are most predictive of neuroanatomy. This is in light of integrated neuromodulator amount being the conventional metric used to monitor neuromodulation in animal studies. Lastly, our study finds that machine learning discrimination of different stimulation strengths or neuroanatomical regions is only possible in adult animals, suggesting a high degree of variability in dopamine modulatory kinetics during animal development. Our study highlights that machine learning could become a broadly utilized tool to differentiate between neuroanatomical regions or between neurotypical and disease states, with features not detectable by conventional statistical analysis.