Network science spans many domains, including computational biology. Biomolecules in the cell do not function alone but instead interact with each other to carry out cellular processes. This is what biological networks model. Efficient computational analyses of biological networks have a potential to deepen our understanding of complex biological processes, which has important biomedical applications, such as identification of disease genes. However, several challenges exist in biological network research: 1) Biological network data are heterogeneous due to availability of various data types. 2) The data represent static snapshots of dynamic cellular functioning. 3) The data are noisy. 4) Many network theoretic problems, such as network comparison and alignment, are computationally intractable. Yet, developing efficient computational strategies to address these challenges is important, as such strategies can assist with e.g., across-species transfer of biological knowledge or protein classification.With these motivations, this Ph.D. dissertation focuses on the development of the following novel computational strategies for integrative, dynamic, and comparative biological network analysis. 1) Since different biological data types can complement each other, we develop a computational framework for integrating different data types to allow for comprehensive data analyses. 2) While current biological network research has been static, to better understand dynamic cellular functioning, we use the integrative framework to infer dynamic biological networks. Then, we analyze the dynamic networks to (computationally) predict new biological knowledge. 3) Since current biological network data are noisy, we computationally de-noise the data to improve the prediction quality. 4) Since network alignment (NA) can aid across-species knowledge transfer, we develop a computational framework for fair evaluation of NA methods. Since a heterogeneous network capturing different data types can allow for more comprehensive data analyses compared to a homogeneous network, and also since traditional NA has been homogeneous, we introduce a novel heterogeneous NA method. Also, we develop a comprehensive framework for alignment-free (as opposed to the above alignment-based) network comparison, which can deal with multiple data types. 5) We use the above approaches to study human aging and for protein classification. Further, we carry out a collaborative application of network-based analyses of the human ubiquitin-proteasome system.