Data-driven approaches for successful biomarker discovery and antibody selection

Proteomics, the large-scale study of proteins in tissues, cells, or fluids, is fundamental to understanding biological processes in health and disease. Protein biomarkers provide critical insights into disease progression and are key to precision medicine. Fluid biomarkers, such as those in blood and cerebrospinal fluid, can enable the early, non-invasive diagnosis of major pathologies including dementias. However, developing reliable biomarkers is challenging, requiring a deeper understanding of molecular properties and antibody interactions for clinical assays. This thesis integrates bioinformatics and machine learning to improve biomarker discovery and antibody selection. It reviews open-access bioinformatics tools and proposes a workflow for integrating them into biomarker assay development. This workflow is illustrated using three established Alzheimer’s disease biomarkers. Two interpretable machine learning models predict key properties relevant to biomarker assays: protein secretion from the brain into the cerebrospinal fluid and protein association with extracellular vesicles. Additionally, the structural analysis of antibody-binding regions enhances our understanding of successful antigen recognition, and a custom R package for immunogen visualization supports antibody selection. A comparative analysis of antibody clustering tools explores relationships between sequence, structure, and function. By addressing challenges such as data bias and model interpretability, bioinformatics tools can further accelerate fluid biomarker development. In summary, this thesis presents multiple computational and data-driven strategies to enhance biomarker discovery and antibody selection.