Until recently, understanding the nature of consciousness was considered a philosophical pursuit. However, technological developments in brain imaging have allowed the study of consciousness as a natural, neurobiological phenomenon. The neurobiology of consciousness has been studied using cognitive and behavioural testing in healthy volunteers and by examining how brain function and connectivity is altered in various clinical settings. The focus of this thesis is to use two of these clinical settings, pharmacologically-induced sedation and disorders of consciousness (DOC), as experimental models for measuring changes in connectivity patterns associated with alterations in consciousness. Experiment 1 presents a method for improving functional magnetic resonance imaging (fMRI) data pre-processing to measure brain network connectivity more accurately. This pre-processing method is then applied to the analyses in the remainder of the thesis. Experiment 2 focuses on a fMRI dataset in which healthy volunteers were administered propofol, an anaesthetic drug known to act on inhibitory GABAergic interneurons. Using a novel multimodal analysis, changes in functional brain network connectivity in default mode, salience, and frontoparietal control networks were found to correlate with the cortical distribution of parvalbumin-expressing GABAergic interneurons. Using the same dataset, Experiment 3 identified a relationship between structural and functional networks in connections between default mode and salience networks. Similar results have been reported in non-human primate models, however, this is the first study to find network-specific structure-function relationships during sedation in humans. These findings informed the remainder of the thesis, which focused on developing network-based machine learning methods for examining brain connectivity in patients with DOC. Experiment 4 developed and validated a graph convolutional neural network (GCNN) classifier using fMRI data and functional connectivity from healthy volunteers performing a volitional mental imagery task. Experiment 5 applied the GCNN to patients with DOC and found frontoparietal control network connectivity measured at rest to be most important in classifying patients capable of performing the mental imagery task. Taken together, these results contribute to the improvement of brain network analysis techniques, the understanding of the neurobiology of propofol-induced sedation, and the development of machine learning algorithms to identify DOC patients with preserved covert volitional capacity. This work demonstrates the utility of clinical models in deepening our understanding of the neurobiology of consciousness.