Abstract:
The brain's remarkable ability to process ambiguous information and transform it into meaningful behavior is a complex process largely performed by neurons. In computational neuroscience, this integration is investigated utilizing large-scale simulations constrained by realistic networks of excitatory and inhibitory neurons. In the cerebral cortex, inhibitory neurons exhibit high variability in cellular properties such as morphology, electrophysiology, and gene expression profiles. This diversity poses a challenge in terms of their characterization and classification across data modalities. While the major subtype specification of inhibitory neurons is given by their molecular identity, it is unknown whether morphological and electrophysiological properties systematically relate to molecular identity to organize the structure and function underlying cortical networks. In this dissertation, I present a computational methodology to assess variations in morphological, electrophysiological, and molecular properties across the entire depth of rat barrel cortex. First, I standardize a comprehensive dataset of morphological and electrophysiological properties, and demonstrate that it is representative for the depth distribution of inhibitory neurons in a cortical column. Then, the molecular composition of the entire rat barrel cortex is acquired, and quantified at 50-micron resolution. For each neuron, I calculate a variety of morphological and electrophysiological features. Multimodal clustering is then utilized to assign neurons into subtypes. Cross-validation with several classifiers is applied to identified subtypes, demonstrating their robustness. The proposed methodology outperforms existing approaches, and its interpretable nature allows me to reliably link different cellular properties across cortical depth. I found that the relative distributions of morphological and electrophysiological properties are similar at any given depth location. At the same time, these distributions systematically shift as a function of cortical depth. Regardless of subtype, the overall axonal and dendritic arborizations, as well as the firing frequency, increase with depth. In contrast, the firing frequency adaptation remains unaffected by depth. Surprisingly, these variations define depth-specific relationships that reveal the molecular identity of inhibitory neurons, which are conserved across species and cortex areas. Thus, simple organizing principles may largely account for the diversity of inhibitory neurons through the adjustment of their morphological and electrophysiological properties to their local environment within cortical circuits, providing novel insight for realistic network modeling.
Machine Learning