AI-driven Biomedical Discovery from Spatial and Single-cell Cancer Data

Genes and their functions can change dramatically in cancerous tissues, often leading to increased disease progression and treatment resistance. Understanding these changes is complex, as gene activity varies between cells and tumor environments. While there is a breadth of scientific literature describing gene functions, relating them to disease mechanisms, and distinguishing causal from correlative relationships is time-consuming and overwhelming for researchers. To address this, we will integrate omics data from spatial assays and single-cell data. This allows visualization of cell locations within tissue and the genes they express. Using data science tools, we will map, compare, and predict roles of genes and cells in tissue contexts across different cancers, e.g., identifying gene signatures indicative of disease-associated tissue properties. We will use Natural Language Processing (NLP) to automatically read and extract knowledge from scientific literature to better understand and inform our omics data analysis, generating hypotheses to explain the differences we see across tissues and cancers. We hypothesise that using the literature will establish causal hypotheses grounded in mechanistic knowledge, leading to more robust and transferable disease signatures. Our work is expected to facilitate interpreting complex gene activity, providing clearer insights, more accurate omics-informed diagnostics, and new avenues for targeted treatment.