Abstract 6753: AI-powered Deep Visual Proteomics (DVP) links pancreatic precursor biology to PDAC survival phenotypes

Abstract Pancreatic ductal adenocarcinoma (PDAC) exhibits 70-80% post-resection recurrence, yet how the earliest molecular events in PanIN precursors connect to divergent survival outcomes remains unclear. We previously established a spatially resolved proteomic atlas of PanINs that revealed molecular reprogramming preceding histological transformation, using Deep Visual Proteomics (DVP). This work mapped acinar-to-ductal metaplasia, incidental ("iPanIN") and cancer-associated ("cPanIN") lesions, and normal ducts into a continuous trajectory to invasive carcinoma and identified four core programs (stress adaptation, immune engagement, metabolic reprogramming, and mitochondrial dysfunction) evident in morphologically normal ducts and low-grade PanINs, alongside cancer-associated field effects. Our MS-based peptide profiling identified KRAS polyclonality at the protein level, independent of genetic sequencing1 .We now extend this approach using tile-level Deep Visual Proteomics (TileDVP2 ) to analyze a first-of-its-kind, multi-institutional FFPE biobank of matched primary-metastatic tumor pairs from short- and long-term PDAC survivors. Using laser microdissection of ∼15 phenotype-matched cells per sample, TileDVP quantifies 3,500-5,000 proteins, enabling high-resolution molecular profiling of tumor cell populations and metastatic niches at the proteomic level. This workflow represents a 300-fold throughput improvement over DVP, while increasing spatial proteomic resolution. We compare proteomic signatures of survival groups across primary tumors, regional and distant metastases, and metastatic niches, and link precursor biology to recurrence patterns, to define how these programs evolve across metastatic sites within and between organs.By unifying precursors, recurrent tumors, and survival phenotypes in a single spatial proteomics framework, this study identifies potential biomarkers for risk stratification and therapeutic targets for earlier intervention in PDAC. 1. Min, J. et al. AI-powered Deep Visual Proteomics reveals critical molecular transitions in pancreatic cancer precursors. bioRxiv 2025.07.07.663528 (2025) doi:10.1101/2025.07.07.663528. 2. Mathian, É. et al. Clinical Image-Based Procedures, 14th International Workshop, CLIP 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025, Proceedings. Lect. Notes Comput. Sci. 21-31 (2025) doi:10.1007/978-3-032-05479-1_3. Citation Format: Jimin Min, Lisa Schweizer, Gijs Zonderland, Benson Chellakkan Selvanesan, Julie H. Thomsen, Lukas Oldenburg, Eduard Chelebian, Benjamin J. Swanson, Michael A. Hollingsworth, Paul Marc Grandgenett, Ishani Ummat, Maximilian Strauss, Andreas Mund, Anirban Maitra. AI-powered Deep Visual Proteomics (DVP) links pancreatic precursor biology to PDAC survival phenotypes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6753.