Abstract B055: A Multimodal Machine Learning Approach for Early Detection of Pancreatic Cancer

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is a deadly cancer, primarily due to late diagnosis and a lack of effective biomarkers. Our prior research has focused on the critical roles of metabolic dysregulation and the tumor microenvironment in the early stages of PDAC. We've shown that RRM1 O-GlcNAcylation disrupts sugar metabolism, leading to genomic instability and the development of oncogenic KRAS mutations. Furthermore, we identified a key signaling pathway: Muc4 overexpression in KrasG12D/+ pancreatic cells promotes fibroblast recruitment and pancreatic intraepithelial neoplasia (PanIN) formation via Activin A secretion. Activin A also mediates tumor-fibroblast interactions that drive metastasis. Importantly, blocking Activin A with Follistatin effectively inhibits PanIN progression and PDAC malignancy in our mouse models. To address the limitations of current early diagnostic methods, we developed a novel multimodal machine learning framework. This integrated strategy combines comprehensive serum metabolomic profiles, clinical information, and the Activin A protein biomarker. This approach demonstrates highly accurate and reproducible early detection of PDAC, offering a promising strategy to improve patient outcomes and with potential for broader applications in cancer diagnostics. Citation Format: Dan-Ni Wu, Joey Jen, Chao-Ping Hsu, Yu-Ting Chang, Chun-Mei Hu. A Multimodal Machine Learning Approach for Early Detection of Pancreatic Cancer [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Pancreatic Cancer Research—Emerging Science Driving Transformative Solutions; Boston, MA; 2025 Sep 28-Oct 1; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2025;85(18_Suppl_3): nr B055.