Gene and Protein Networks in Understanding Cellular Function

Over the past decades, networks have emerged as a useful way of representing complex large-scale systems in a variety of fields. In cellular and molecular biology, gene and protein networks have attracted considerable interest as tools for making sense of increasingly large volumes of data. Despite this interest, there is still substantial debate over how to best exploit network models in cellular biology. This thesis explores the use of gene and protein networks in various biological contexts. The first part of the thesis (Chapter 2) examines protein function prediction using network-based ‘guilt-by-association’ approaches. Given the falling costs of genome sequencing and the availability of large volumes of biological data, automated annotation of gene and protein function is becoming increasingly useful. Chapter 2 describes the development of a new network-based protein function prediction method and compares it to a leading algorithm on a number of benchmarks. Biases in benchmarking methods are also explicitly explored. The second part (Chapters 3 and 4) explores network approaches in understanding loss of function variation in the human genome. For a number of genes, homozygous loss of function appears to have no detrimental effect. A possible explanation is that these genes are only necessary in specific genetic backgrounds. Chapter 3 develops methods for identifying these types of relationships between apparently loss of function tolerant genes. Chapter 4 describes the use of networks in predicting the functional effects of loss of function mutations. The third part of the thesis (Chapters 5 and 6) uses network representations to model the effects of cellular stress on yeast cells. Chapter 5 examines stress induced changes in co-expression and protein interaction networks, finding evidence of increased modularisation in both types of network. Chapter 6 explores the effect of stress on resilience to node removal in the co-expression networks