In this project, we analyzed the clinical MEMORI trial, which investigated a therapy switch of early PET (Positron emission tomography) non-responders after induction chemotherapy to chemoradiation in locally advanced gastroesophageal adenocarcinomas (GEJ Ca). The primary endpoint to increase the R0 resection rate in metabolic non-responders was not met. PET response after induction therapy was prognostic for outcome with a prolonged survival in PET responders. We explored novel PET imaging parameters and used deep learning to predict chemotherapy response from histopathology images, which supported development of a transformer-based network tool (CellViT) for automated segmentation of cell nuclei in digitized tissue samples. Multi-omics analysis of biopsies before, during and after therapy revealed profound transcriptional changes during treatment with limited evidence of clonal replacement, suggesting cell plasticity and immune dynamics as key mechanisms. The MEMORI project built on advances of the German Cancer Consortium (DKTK), enabling consistent biobanking and high-quality data acquisition across multiple sites and supported career development of young researchers.