Interpretable extracellular vesicle long RNA framework for noninvasive pancreatic cancer diagnosis: A multi-omics artificial intelligence-driven liquid biopsy paradigm

Figure 1 Conceptual workflow of the interpretable two-layer machine learning framework (ECD-itMLF) for pancreatic ductal adenocarcinoma diagnosis. Created with Picdoc. A: Plasma extracellular vesicle isolation and long RNA sequencing; B: Dimensionality reduction layer integrating singular value decomposition, nonlinear iterative partial least squares, and probabilistic principal component analysis to construct the extracellular vesicle long RNA-index; C: Two-layer classification strategy with support vector machine screening followed by random forest differentiation; D: Biological interpretability through pathway enrichment and single-cell correlation analysis. EV: Extracellular vesicle; EV-lRNA: Extracellular vesicle long RNA; SVD: Singular value decomposition; NIPALS: Nonlinear iterative partial least squares; PPCA: Probabilistic principal component analysis; SVM: Support vector machine; PDAC: Pancreatic ductal adenocarcinoma; EMT: Epithelial-mesenchymal transition; TGF-β: Transforming growth factor-β.

Figure 2 Future multi-omics integration strategy for pancreatic ductal adenocarcinoma liquid biopsy. Created with Picdoc. The integration of extracellular vesicle long RNA, circulating tumor DNA mutations, extracellular vesicle-derived proteins (GPC1 and CD63), and metabolic profiles enables comprehensive molecular characterization. Artificial intelligence-driven fusion of these layers, combined with radiomic features from medical imaging, promises enhanced diagnostic accuracy and minimal residual disease monitoring. EV: Extracellular vesicle; EV-lRNA: Extracellular vesicle long RNA; ctDNA: Circulating tumor DNA; AI: Artificial intelligence; MRD: Minimal residual disease.