Abstract Pancreatic ductal adenocarcinoma (PDAC) is characterized by relatively low tumor purity and an abundant tumor microenvironment. To dissect the contribution of the biologic components, we developed DECODER, which performs de novo compartment deconvolution and weight estimation of tumor samples. DECODER is a sophisticated framework that integrates runs of non-negative matrix factorization (NMF) and non-negative least square (NNLS) algorithms and can be applied to any non-negative matrices without the need to know the number of resultant factors or compartments. DECODER was used to deconvolve the TCGA pancreatic adenocarcinoma (PAAD) RNA-seq dataset, which resulted in the identification of 7 major compartments (basal tumor, classical tumor, activated stroma, normal stroma, immune, endocrine, and exocrine), confirming prior manual NMF-based solutions. These results were then used for single-sample based weight estimation in the COMPASS trial and ICGC PACA-AU RNA-seq dataset. We saw a significant positive correlation between DECODER immune weight and leukocyte fraction (r = 0.757, p < 0.001) or ESTIMATE immune score (r = 0.773, p < 0.001). Samples with high immune weights corresponded to immune infiltration by histology. A significant correlation was found between the sum of basal and classical tumor weights, and tumor purity based on both ABSOLUTE (r = 0.699, p < 0.001) and methylation (r = 0.71, p < 0.001). Similarly, the sum of activated and normal stroma weights correlated with ESTIMATE stromal score (r = 0.729, p < 0.001). Interestingly, we found that the ratio between the basal and classical compartment (bcRatio) was significantly associated with survival outcome (p = 0.049 in TCGA and 0.008 in ICGC) in all patients and treatment response in basal-like patients (r = 0.884, p < 0.001 in COMPASS trial), suggesting that bcRatio may help explain the molecular basis for tumor behavior in PDAC. DECODER was also applied for de novo deconvolution for all the cancer types in TCGA RNA-seq dataset and identified the cancer type specific compartments. Results from DECODER can then be used for single-sample weight estimation of new samples for any cancer type. In addition, we applied DECODER on the PanCan ATAC-seq dataset containing 23 cancer types in a combined fashion, and identified compartments associated with cancer types or organ systems. This proves that DECODER is highly feasible to data of multiple platforms. In summary, we present an automated method for de novo deconvolution that may be used across tumor and data types. With deconvolved results as the reference, DECODER enables the single-sample weight estimation for a new sample, which is plausible in the clinical setting. Citation Format: Xianlu L. Peng, Richard A. Moffitt, Robert J. Torphy, Keith E. Volmar, Jen Jen Yeh. Compartment deconvolution in pancreatic cancer with biologic and clinical implications [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer: Advances in Science and Clinical Care; 2019 Sept 6-9; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2019;79(24 Suppl):Abstract nr B41.