Modelling and investigation sleep / deep sleep in Drosophila

Sleep is a broadly conserved phenomenon that is thought to be vital due to its severe disadvantages, such as being less responsive and in an inactive state. Drosophila Melanogaster has emerged over the last two decades as a new tool to investigate the genetics and behaviour of sleep after seminal papers from Hendricks et al. and Shaw et al. demonstrated that immobility > 5 minutes met the criteria for a sleep-like state. However, despite the rule being proven robust generally, it is likely not accurate for all cases such as mutant variants, sex, and species, given its inflexibility.

To this extent, I will be developing a machine learning model to analyse and predict sleep in Drosophila. Machine learning tools excel in clustering and classifying large inputs of data relatively free from human bias and arbitration. Sleep will be predicted from movement data recorded in an ethoscope, a set of hardware and software utilising machine vision to track Drosophila movement in a contained tube that was developed in the Gilestro Labs. The ethoscope records movement and immobility at a higher rate than traditional beam break experiments and thus makes it prime for data rich machine learning. Due to the sequential nature of the data I will be initially applying the frameworks of Hidden Markov Models (HMM) and recurrent neural networks (RNN) initially. The data will be trained and applied to a variety of Drosophila to better understand differences in sleep. Suitable model predictions will be checked in ground truth experiments, comparing the predictions against the traditional 5-minute rule and arousal threshold data, which will entail response to both air and favourable smells. Additionally, I will look to expand the model to predict deep sleep in Drosophila, a contested sleep state that is said to be associated with an increased arousal threshold above 'normal sleep'. If deep sleep occurs in Drosophila the model if tuned properly should detect a difference in the pattern of sleep that correlates with an increasing arousal threshold.

Additionally, I will look to explore the RNA profile of chronically sleep deprived Drosophila. The ethsocope can deprive flies of sleep through mechanical stimulation after a set time of immobility, to create sleep deprivation at a far greater level than previously seen. I will compare the neuronal expression profile of the flies from acute to chronic sleep deprivation to identify novel upregulated and down regulated genes. Suitable candidates will be taken forward for expression profiling to identify where it is expressed and its surrounding interaction network.

Together these projects should strengthen the robustness of Drosophila as a model organism for sleep and generate greater insight into the molecular and genetic mechanisms underlying sleep and sleep deprivation.