Background: Pancreatic cancer remains an extremely malignant digestive tract tumor, posing a significant global public health burden. Patients with pancreatic cancer, once metastasis occurs, lose all hope of cure, and prognosis is extremely poor. It is important to investigate liver metastasis of Pancreatic cancer in depth, not just because it is the most common form of metastasis in pancreatic cancer, but also because it is crucial for treatment planning and prognosis assessment. This study aims to delve into the mechanisms of pancreatic cancer liver metastasis, with the goal of providing crucial scientific groundwork for the development of future treatment methods and drugs.Methods: We explored the mechanisms of pancreatic cancer liver metastasis using single-cell sequencing data (GSE155698 and GSE154778) and bulk data (GSE71729, GSE19279, TCGA-PAAD). Initially, Seurat package was employed for single-cell data processing to obtain expression matrices for primary pancreatic cancer lesions and liver metastatic lesions. Subsequently, high-dimensional weighted gene co-expression network analysis (hdWGCNA) was used to identify genes associated with liver metastasis. Machine learning algorithms and COX regression models were employed to further screen genes related to patient prognosis. Informed by both biological understanding and the outcomes of algorithms, we meticulously identified the ultimate set of liver metastasis-related gene (LRG). In the study of LRG genes, various databases were utilized to validate their association with pancreatic cancer liver metastasis. In order to analyze the effects of these agents on tumor microenvironment, we conducted an in-depth analysis, including changes in signaling pathways (GSVA), cell differentiation (pseudo-temporal analysis), cell communication networks (cell communication analysis), and downstream transcription factors (transcription factor activity prediction). Additionally, drug sensitivity analysis and metabolic analysis were performed to reveal the effects of LRG on gemcitabine resistance and metabolic pathways. Finally, functional experiments were conducted by silencing the expression of LRG in PANC-1 and Bx-PC-3 cells to validate its influence to proliferation and invasiveness on PANC-1 and Bx-PC-3 cells.Results: Through a series of algorithmic filters, we identified PAK2 as a key gene promoting pancreatic cancer liver metastasis. GSVA analysis elucidated the activation of the TGF-beta signaling pathway by PAK2 to promote the occurrence of liver metastasis. Pseudo-temporal analysis revealed a significant correlation between PAK2 expression and the lower differentiation status of pancreatic cancer cells. Cell communication analysis revealed that overexpression of PAK2 promotes communication between cancer cells and the tumor microenvironment. Transcription factor activity prediction displayed the transcription factor network regulated by PAK2. Drug sensitivity analysis and metabolic analysis revealed the impact of PAK2 on gemcitabine resistance and metabolic pathways. CCK8 experiments showed that silencing PAK2 led to a decrease in the proliferative capacity of pancreatic cancer cells and scratch experiments demonstrated that low expression of PAK2 decreased invasion capability in pancreatic cancer cells. Flow cytometry reveals that PAK2 significantly inhibited apoptosis in pancreatic cancer cell lines. Molecules related to the TGF-beta pathway decreased with the inhibition of PAK2, and there were corresponding significant changes in molecules associated with EMT.Conclusion: PAK2 facilitated the angiogenic potential of cancer cells and promotes the epithelial-mesenchymal transition process by activating the TGF-beta signaling pathway. Simultaneously, it decreased the differentiation level of cancer cells, consequently enhancing their malignancy. Additionally, PAK2 fostered communication between cancer cells and the tumor microenvironment, augments cancer cell chemoresistance, and modulates energy metabolism pathways. In summary, PAK2 emerged as a pivotal gene orchestrating pancreatic cancer liver metastasis. Intervening in the expression of PAK2 may offer a promising therapeutic strategy for preventing liver metastasis of pancreatic cancer and improving its prognosis.

Background This study aims to explore a deep learning (DL) algorithm for developing a prognostic model and perform survival analyses in SBT patients. Methods The demographic and clinical features of patients with SBTs were extracted from the Surveillance, Epidemiology and End Results (SEER) database. We randomly split the samples into the training set and the validation set at 7:3. Cox proportional hazards (Cox-PH) analysis and the DeepSurv algorithm were used to develop models. The performance of the Cox-PH and DeepSurv models was evaluated using receiver operating characteristic curves, calibration curves, C-statistics and decision-curve analysis (DCA). A Kaplan–Meier (K–M) survival analysis was performed for further explanation on prognostic effect of the Cox-PH model. Results The multivariate analysis demonstrated that seven variables were associated with cancer-specific survival (CSS) (all p < 0.05). The DeepSurv model showed better performance than the Cox-PH model (C-index: 0.871 vs. 0.866). The calibration curves and DCA revealed that the two models had good discrimination and calibration. Moreover, patients with ileac malignancy and N2 stage disease were not responding to surgery according to the K–M analysis. Conclusions This study reported a DeepSurv model that performed well in CSS in SBT patients. It might offer insights into future research to explore more DL algorithms in cohort studies.

Background: Pancreatic ductal adenocarcinoma (PDAC) remains among the most lethal solid tumors, largely due to its intricate and immunosuppressive tumor microenvironment (TME). While single-cell sequencing technologies have begun to unravel the cellular heterogeneity of PDAC, a comprehensive understanding of how genetic determinants influence and are influenced by the TME is still lacking. To bridge this knowledge gap, our study employs an integrated multi-omics approach, incorporating single-cell transcriptomics, genomics, and proteomics, complemented by computational biology and machine learning. We aimed to delineate the core molecular drivers of PDAC pathogenesis, with subsequent in vitro functional validation focusing on the role of UNC93B1 in malignant phenotypes. The ultimate goal of this research is to inform the development of precise therapeutic strategies to enhance patient survival and quality of life. Methods: We assembled a comprehensive multi-omics dataset, including single-cell RNA-seq data from 22 PDAC samples (GSE154778, GSE212966), bulk transcriptomic cohorts (GSE28735, GSE62452), survival data from the TCGA-PAAD project (n=172), spatial transcriptomics, and genome-wide association study data (bbj-a-140, n=196,187). The single-cell data were processed using Seurat v5, which involved rigorous quality control, batch effect correction with Harmony, unsupervised clustering, and cell type annotation to characterize TME heterogeneity. Genetic susceptibility was mapped onto single-cell data using scPagwas to calculate trait-regulated scores (TRS) and identify trait-associated genes. Co-expression networks were constructed via high-diversity WGCNA (hdWGCNA), and key candidate genes were refined through survival analysis and a machine learning framework integrating LASSO regression, Random Forest, and Support Vector Machine algorithms. The functional role of the pivotal gene, UNC93B1, was systematically investigated through Gene Set Variation Analysis (GSVA), pseudotime trajectory inference (Monocle2), and cell-cell communication analysis (CellChat). In vitro validation was performed using four PDAC cell lines (PANC-1, BxPC-3, Capan-1, SW1990). Following qPCR confirmation of high UNC93B1 expression, a stable knockdown model (sh-UNC93B1) was generated in Capan-1 cells. Functional consequences were assessed using CCK-8, wound healing, transwell and colony formation assays. A subcutaneous xenograft model was established to evaluate tumor growth in vivo . Mechanistic insights were gained through flow cytometry for cell cycle analysis and molecular profiling of the cGAS-STING pathway, senescence markers (e.g., p16^INK4a), and epithelial-mesenchymal transition (EMT)-related genes. Results: Single-cell transcriptomic profiling delineated nine distinct cell populations within the PDAC TME. hdWGCNA identified three gene modules (8, 11, 16) positively associated with tumorigenesis. The intersection of these modules with differentially expressed genes yielded 320 candidates, which were subsequently filtered to 61 genes significantly linked to patient prognosis (P < 0.05) via Cox regression. Cross-validation across machine learning models and scPagwas analysis converged on UNC93B1 as the sole overlapping gene with consistent diagnostic and prognostic relevance. UNC93B1 was robustly upregulated in tumor tissues across independent datasets (TCGAxGTEx, bulk RNA-seq), a finding corroborated at the protein level by HPA and CPTAC data (P < 0.01). Its expression positively correlated with higher pathological grade and was spatially enriched within tumor regions. Functional enrichment analysis (GSVA) suggested that UNC93B1 is involved in the suppression of the cGAS-STING signaling axis. Pseudotime analysis indicated that UNC93B1 expression escalates along tumor progression trajectories. CellChat suggested strengthened intercellular communication networks in UNC93B1-high cells, particularly modulated by the cGAS-STING pathway. In vitro , UNC93B1 knockdown in Capan-1 cells significantly attenuated proliferative capacity (22.3% reduction in OD450 at 72h, P < 0.05), migratory ability (29.6% reduction in wound closure, P < 0.05), and clonogenic survival (342 fewer colonies, P < 0.01). Mechanistically, sh-UNC93B1 cells exhibited G1/S phase arrest (8.9% increase, P < 0.05), activation of the STING/IFN-β/CXCL10 cascade, elevated p16^INK4a expression, and a reversal of EMT, evidenced by downregulation of VIM and upregulation of CDH1. Consistently, in vivo xenograft experiments demonstrated that UNC93B1 silencing markedly impeded tumor growth, concomitant with reduced UNC93B1 protein and enhanced STING pathway activation. Critically, the tumor-suppressive phenotypes induced by UNC93B1 knockdown, including the inhibition of proliferation, migration, and clonogenicity, were largely reversed upon treatment with the selective STING inhibitor H-151, confirming that the observed functional consequences are causally mediated through the activation of the cGAS-STING pathway. Conclusion: By integrating multi-omics data, including GWAS and spatial transcriptomics, this study systematically defines a pivotal role for UNC93B1 in PDAC progression. Our findings demonstrate that UNC93B1 is associated with an immunosuppressive TME and facilitates metastatic spread, potentially through inhibiting the cGAS-STING-mediated innate immunity pathway. The strong correlation between UNC93B1 overexpression and adverse clinical outcomes underscores its potential as a dual diagnostic biomarker and therapeutic target. This work not only provides a mechanistic foundation for novel precision immunotherapies in PDAC but also establishes a robust methodological paradigm for multi-omics-driven discovery in oncology.