Computational Network Analytics for Applications in Biomedicine

Network medicine has recently emerged as a field of research focusing on the analysis of networks modelling complex biological systems, for a better understanding of diseases and corresponding treatment. Building on results from graph theory, it provides a network-oriented approach for the identification of potential points of interest in these systems. Within this context, diseases can be regarded as systemic dysregulations in a patient’s specific interaction network, while drug therapeutics represent the external interventions aiming to offset the effects of the disease. The disease data, which can include disease-drivers, typical genetic and functional dysregulations, or prospective drug and drug-target details, can be integrated into comprehensive networks that can help with the identification of targeted drugs and combinations thereof. There are multiple approaches to the study of these networks, such as through topological analysis or time-based dynamics. The recent availability of high-quality biological data and improvements in algorithmics and computational techniques reinforce the strong potential of the methods and their immediate applicability in the biomedical domain.

The first part of the thesis focuses on network controllability, which pertains to the ability to guide a network to a desired state through minimal external interventions through the identification of nodes of interest within. We provide a brief theoretical background for the structural controllability problem and several of its approaches, such as target- or input-constrained structural controllability, together with an overview of the existing algorithms and tools and followed by a short a discussion on the necessity of developing approachable software implementations for both existing and novel efficient algorithms. We prove that the target variant of the problem is hard to approximate and fixed-parameter tractable, and we introduce several algorithms aimed at solving it: an exhaustive search algorithm bounded by naturally constrained limits, and an approximation genetic algorithm which uses an algebraic approach. Moreover, we extend these algorithms to efficiently solve the input-constrained variant of the problem.

The second part of the thesis focuses on the applicability of the structural controllability approaches in biomedicine. We talk about the generation of personalized protein-protein interaction network around disease-, patient-, or drug-specific proteins of interest, and their subsequent analysis, together with possible interpretations for the results. We apply this framework for the identification of potential repurposable drug suggestions for COVID-19 and breast, ovarian, and pancreatic cancer, as well as for the suggestion of personalized treatment for three multiple myeloma patients. Following these case studies and motivated by the lack of dedicated tools and the multitude of available biological databases, we introduce a novel, free, and open-source web-based platform allowing for the generation and structural controllability analysis of customized protein-protein interaction networks. This novel and, to the best of our knowledge, unique software integrates the controllability algorithms and multiple biological data sources and aims to enable direct, easy, and widespread usage of the presented methods.

This thesis focuses on developing a framework for the application of structural controllability in biomedicine. The work encompasses theory, data, algorithms, and their implementation for the identification of potential novel drugs and drug combination suggestions, drug-repurposing candidates, and treatment lines, moving forward towards personalized approaches to therapeutics.