Abstract A071: Deep learning enabled 3D multi-omic analysis reveals molecular signatures of heterogeneous response to chemotherapy in pancreatic cancer

Abstract

Resistance to systemic therapy is a major unmet challenge in pancreatic cancer. To identify potential mechanisms of resistance, we developed a novel 3D pipeline in clinical samples that uses deep learning to classify sensitive and persistent tumor cell populations based on morphological features, enabling subsequent molecular characterization of intratumoral heterogeneity. We applied this automated 3D pipeline to a cohort of human pancreatic cancer samples treated with neoadjuvant chemotherapy, identifying heterogeneity in response to therapy both between and within tumors. Application of spatial proteomics to these sensitive and persistent regions identified enhanced epithelial-to-mesenchymal transition and non-classical cell states in persistent cells, confirming our morphological classification. Integration of spatial transcriptomics in multiple pancreatic cancer cohorts associated fibroblast-cancer crosstalk via syndecans with resistance to cytotoxic therapy. Our validated 3D multi-omic pipeline is now poised for application to clinical trials, enabling discovery of resistance mechanisms and design of new therapeutic combinations to circumvent resistance.