Large-scale modeling and simulation of neuronal networks in the human cortex

Neuroscience is concerned with how a complex network of remarkably intricate neurons and synapses gives rise to cognition. The cerebral cortices of macaque and human are studied in this thesis and consist of billions of neurons and trillions of synapses. Due to the sheer number of cells, experimental studies can only focus on a limited number of cells and locations simultaneously. However, in theory, computational studies are not limited by the size of the network: The upper bound is imposed by the size and power of the computer and the performance of the simulation software. Here, we study how bottlenecks in simulation software can be overcome and how biological data can be used to build large-scale computational models of cortices, enabling the study of dynamics and function. We identify the delivery of spikes to target neurons as especially time-consuming in simulations of large-scale models. This phase involves inherent random access to memory that cannot easily be alleviated and causes long latencies. We describe how this phase can be rearranged such that interleaved function calls mask the latencies. The intricacies of this study expose the need for a systematic benchmarking framework. In light of multiple computers, simulation codes, and neuroscientific models, it is imperative to keep track of results and metadata. We present a framework that eases the challenges involved with benchmarking. To study the link between cortical structure and the dynamics of brain networks, we integrate biological data on connectivity and cytoarchitecture into a model of human cortex. The model is layer- and population-resolved and statistical regularities fill gaps in the experimental data. We poise the model into a state where the activity matches single-cell cortical recordings and whole-brain functional magnetic resonance imaging scans. We study how a perturbation in the form of a single spike propagates through the network. The following study takes this a step further: We study the effect of clustering of neuronal ...