PROJECT SUMMARY Fully characterizing genomic changes is a central challenge in understanding biological processes and creating effective treatment targets for complex, multi-factorial diseases like cancer. Genomic and epigenomic data have been used to subcategorize pancreatic ductal adenocarcinoma (PDAC) into more meaningful and functionally important groups. This data can be used to understand unique biological and molecular tumor signatures for treatment and improve survival. Typical multi-omics approaches use single-feature association analysis that requires the feature to have the same behavior in all patients in each group and high homogeneity in each subgroup. We propose a more realistic analysis using the convolution neural network (CNN) deep learning approach to analyze multi-omics datasets. CNN is designed to identify subsets of features important in subsets of patients. We will enrich existing PDAC omics datasets developed in our previous research collaboration, where an omics-based approach was used to identify the PDAC-associated alterations across omic sets (e.g., expression, methylation, sequence, etc.). We propose identifying the functionally important genomic features associated with core gene (KRAS, SMAD4, and TP53) mutations that lead to PDAC progression and survival. We will identify high-leverage biomarker/treatment targets with the following study objectives. (1) Assess and refine core gene involved sufficient-component survival models for participants with paired normal and tumor tissues, (2) use genetically engineered mouse models (KPC and PKS) to refine these models, and (3) train deep learning methods to identify gene mutations, epigenetic, and clinical features associated with the progression and survival of PDAC tumors. Our team is ideally suited to accomplish these aims, as we have developed bioinformatics workflows and conducted epigenetic and genome-wide analyses in multiple disease conditions using numerous study types in humans and mice. This proposed work will use the CNN deep learning method to explore multi-omics data and validate combinations of clinical, genetic, and epigenetic features associated with PDAC progression and survival previously observed by our group. We will use genomic feature-defined subtypes of PDAC and incorporate clinical information, and the methods and framework used in this project can be applied to other cancers. The proposed work will reveal functionally meaningful relationships between genomic features (e.g., methylation, miRNA, lncRNA, ATAC, core gene mutations) and core gene expression in PDAC tissue and adjacent normal samples. This will provide targets for drug development and has the potential to benefit millions of people. This work will also lay the foundation for integrating genomic data within the clinic to provide individualized PDAC tumor treatment plans.