Computational studies of protein interactions and genetic regulation

The work in this thesis is split into two parts. The introduction and following two chapters pertain to the investigation of gene regulation using Chip-seq data and linear modelling. The final chapter pertains to the prediction of hot-spot residues in protein-protein interactions. The rapid escalation in the speed and quality of DNA sequencing has lead to a wealth of data for the location of transcription factor binding and histone modifications across the genomes. Using Chromatin ImmunoPrecipitation followed by sequencing (ChIP-seq) data, we have generated a new binding metric based on the enrichment of the read-counts for each gene. Eight datasets from mouse macrophage cells (two histone modifications, five transcription factors, DNase I hypersensitivity) were used to model the binding of RNA polymerase II. It was found that a linear model just using the DNase I hypersensitivity and histone modification data was better than any of the models containing the transcription factor data. Investigation of the outlying genes for the model revealed no pattern in their Gene Ontology terms or macrophagespecific genes. Human embryonic stem cell data (23 transcription factor and 24 histone modification datasets) were used in combination with LASSO regression to model the binding of RNA polymerase II. The resultant models contrasted with the results from the mouse macrophage linear models in that using the histone modifications data in combination with the transcription factor data lead to the best models. A much more complicated picture of the regulation of RNA polymerase II binding was produced using the LASSO models. Protein-protein interactions are essential for every function within a cell and being able to predict them has large consequences for drug discovery and understanding the vast proteininteraction networks that occur within cells. Predicting protein-protein interactions is difficult due to the large number of possible conformations; predicting hot-spot residues can greatly reduce this. InterBasePro was compared with experimental data and subsequently adaptation was done to assess its usefulness for predicting hot-spot residues. An alternative approach was also made into classifying hot-spot residues based on atomic contacts.