Computational methods for the analysis of non-cell-autonomous phenomena and derived gene co-expression networks

Abstract

Non-cell-autonomous effects are the changes observed in one cell or cell-type as a consequence of the actions of another. The study of these phenomena is crucial to our understanding of how diverse cell-types function and co-operate together in complex tissues. The investigation of these effects has been greatly advanced by the advent of next-generation sequencing (NGS) technologies which enable the rapid sequencing of genetic information. NGS data, such as RNA-Seq, can be analysed computationally to allow comparison of cellular transcriptomes. In practice, the study of non-cellautonomous phenomena through NGS has relied upon the physical separation of cell populations in order to be sure that derived transcriptomic data is exclusively from one cell type or the other. However these methods have been shown to introduce noise as a result of stress induced by the separation process, whilst also being susceptible to bias through contamination resulting from imperfect separation of cell populations. In this thesis, a pipeline was developed to provide an in silico means of investigating these phenomena without the need for physical separation. The pipeline takes RNA-seq reads from novel mixed-species populations - in vitro cultures where each cell type is derived from a distinct species - and sorts them according to species specific origin using quality variables from multiple genome mappings as discriminators. Our method is demonstrably robust to incorrect assignment and shows high precision and recall across species of differing genetic distances, thereby providing an alternative to flawed physical separation techniques. Downstream study of such RNA-seq samples is increasingly conducted using network methodologies. Gene co-expression networks have been demonstrated as a biologically representative means for analysing NGS data. However, many existing methods for attributing the involvement of biological function to networked datasets disregard the structural information provided within them. In this thesis, I build upon an existing approach to use information theoretic entropy as a method for network-based enrichment and thereby demonstrate that the integration of network edge information can be used to more reliably infer biological pathway involvement. Our method out-performs the original whilst correcting for pathway-size bias. Lastly, the utility of the methods presented in this thesis was demonstrated through application to the study of two different phenomena: the induction of neural activity on co-cultures of neurons with astrocytes and the stimulation of microglia by LPS on co-cultures of microglia, neurons and astrocytes, by investigating cell-type specific involvement of biological pathways.