Bioinformatics is grappling with an explosion of data, creating both opportunities and challenges for scientific discovery and healthcare. This thesis stands at the crossroads of systems medicine and privacy-aware artificial intelligence (AI), offering contributions that aim to harness the potential of this data-rich landscape. Central to the thesis are the web tools CoVex, sPLINK, FeatureCloud, and AIMe, each designed to address unique challenges, publicly and freely available to the research community.
Within the ambit of systems medicine, CoVex emerges as a tool in the realm of infectious diseases and drug repurposing. Deploying network exploration and ranking algorithms like centrality measures, CoVex identifies intricate disease pathways and potential drug targets. Its purpose mainly lies in drug repurposing achieved by its capability to explore integrated virus-protein, protein-protein and protein-drug interaction networks to identify alternative applications for existing drugs, thereby accelerating the medical response to urgent challenges like the COVID-19 pandemic.
Privacy-aware AI is the second major pillar of the thesis, with a focus on federated learning (FL) as an enabling technology. The tools sPLINK and FeatureCloud are introduced to demonstrate this approach. sPLINK, specialized for genome-wide association studies (GWAS), preserves data privacy without compromising analytical robustness. FeatureCloud expands upon this by serving as a versatile, FL platform, thereby facilitating large-scale analyses across multiple institutions while adhering to stringent data privacy norms. It employs and integrates state-of-the-art privacy-enhancing techniques (PETs), such as differential privacy (DP) and secure multiparty computation (SMPC), to protect sensitive patient data. Evaluation of FeatureCloud shows that the results are sufficiently close or even identical to centrally performed analyses, thereby demonstrating the efficacy and applicability of FL in a cross-silo context.
The thesis also brings forth the AIMe registry, aiming to create a foundation for transparency, reproducibility, and reliability in biomedical AI. By setting standards and ensuring correct and complete reporting, AIMe acts as a central hub for vetting and disseminating AI tools, increasing validation and reproducibility of results reported in biomedical research.
As we traverse an era defined by rapid data proliferation and stringent data protection laws, this thesis demonstrates that specialized tools and versatile platforms are valuable additions to the research landscape. CoVex, sPLINK, AIMe and FeatureCloud each have unique specializations, yet they all contribute to more efficient research in systems medicine: making integrated data quickly accessible to researchers, allowing large-scale analyses across distributed datasets, and ensuring valid and reproducible reporting of results.