Automated cancer detectors based on neural networks streamline diagnosis and prognosis by enhancing accuracy and efficiency, reducing the manual effort of professional pathologists and speeding up the analysis process. However, digitised microscopy scans of tissue samples are inappropriate for direct model training due to their vast size. A typical approach is to cut the images into much smaller patches, resulting in the loss of relational information. To overcome this, graph neural networks can be used as they maintain the contextual relationships between the tiles. This thesis aims to design, train and evaluate a graph neural network for prostate cancer prediction using data and annotations provided by the Masaryk Memorial Cancer Institute. The student will familiarise themself and present graph neural networks and the basics of digital pathology, focusing on prostate cancer diagnosis. They will discuss and select suitable architecture and perform and asses a comparative test in the cloud infrastructure of the national group. The student will optionally contribute the code to HistoPipe, the RationAI open-source library for digital histopathology.