Xiujie Chen

Gastric Cancer Stem Cells (GCSCs) contribute to tumor progression and treatment resistance, but their heterogeneity and tumor microenvironment interactions remain unclear. We analyzed GSE163558 single-cell RNA-sequencing data from 10 gastric cancer samples and integrated GCSC-related genes with The Cancer Genome Atlas Stomach Adenocarcinoma (TCGA-STAD) training cohort (n = 448) and two independent external validation cohorts, GSE62254 (n = 300) and GSE84437 (n = 483), to construct and validate a six-gene prognostic risk model. Cell–cell communication, metabolic activity, immune infiltration, immunotherapy response, and drug sensitivity were evaluated. Nine cell populations, including a putative GCSC population, were identified. GCSCs showed higher inferred oxidative phosphorylation activity and extensive predicted interactions with tumor-associated cells. The model stratified patients into distinct prognostic groups. High-risk tumors exhibited enrichment of progression-related pathways, including epithelial–mesenchymal transition, and distinct immune features. In PRJEB25780 (n = 78), higher risk scores were associated with immunotherapy non-response, whereas an exploratory analysis of GSE91061 (n = 49) showed a similar trend. The model may support prognostic stratification and provide insights into potential associations with therapeutic response, pending prospective validation.

Introduction: Oral squamous cell carcinoma (OSCC) is characterized by inter- and intratumoral heterogeneity, which complicates patient stratification and treatment. Although single-cell studies have improved the resolution of the OSCC microenvironment, the relationships among cellular states, spatial organization, immune-repertoire dynamics, and patient-level ecosystem patterns remain incompletely understood. Methods: We integrated single-cell RNA sequencing, single-cell T-cell receptor sequencing, spatial transcriptomics, and bulk RNA-sequencing data to characterize OSCC. Results: We constructed a multi-omics cellular atlas comprising 88,442 cells and 63 cell subclusters, revealing complementary malignant-cell heterogeneity at the levels of transcriptional subclusters and recurrent meta-programs. Spatial and cell-cell communication analysis identified a reproducible association between PI3+ cancer-associated fibroblasts and APC_Myeloid cells and supported a candidate collagen-associated communication axis within tumor niches. Integration of paired transcriptomic and TCR data delineated a CD8+ T-cell pseudotime continuum accompanied by increasing clonal expansion and exhaustion-associated features, with spatial analysis linking intermediate trajectory states to regulatory T-cell and inflammatory NKT-cell-rich tumor-front regions. Extending the single-cell atlas to bulk cohorts by deconvolution, we defined four OSCC ecotypes and four coordinated cell programs that captured patient-level ecosystem organization and showed distinct molecular and clinical associations. We developed exploratory models for prognosis, recurrence, and immunotherapy sensitivity and established the OSCC Atlas as an online resource for data exploration. Discussion: This integrative framework links cellular heterogeneity, spatial organization, immune dynamics, and patient-level ecology, providing testable hypotheses for the biological and clinical stratification of OSCC.

Chronic infection with Helicobacter pylori (H. pylori) is a major environmental risk factor for gastric carcinogenesis. Malignancy is largely driven by variations within the virulence factor CagA, with East Asian lineages exhibiting higher oncogenic potential than Western ones. However, how these variants modulate cellular crosstalk remains poorly understood. We integrated molecular dynamics (MD) simulations with single-cell transcriptomics across progressive disease stages, including chronic atrophic gastritis, intestinal metaplasia, and gastric cancer. Local niche remodeling was evaluated via cell–cell communication profiling among epithelial, stromal, and immune circuits, while simulations of MARK2 kinase bound to distinct CagA lineages determined binding affinities. Single-cell analysis revealed that H. pylori toxicity progressively dampens epithelial–stromal crosstalk, marked by severe epithelial polarity aberrations that disrupt neuroendocrine-like secretory and synaptic pathways during malignant transformation. Mechanistically, MD simulations and MM/GBSA calculations demonstrated that East Asian CagA lineages exhibit higher binding affinity toward host MARK2 than Western lineages. Specific East Asian amino acid substitutions dramatically tighten the protein interface, driving stronger signaling perturbations. This study bridges atomistic structural virulence with microenvironmental shifting, establishing geographic CagA toxicity divergence as a critical determinant for pathogen-driven gastric cancer risk.

The leading-edge (LE) of hepatocellular carcinoma (HCC) is a critical region driving malignant progression and is closely associated with high patient mortality and marked intratumoral heterogeneity. Multi-omics integration identified elevated expression of SPARC and IGFBP7 in the LE region, which was associated with stromal remodeling-related transcriptional programs and an immune-depleted microenvironment. Cell-cell communication and pathway analyses further suggested potential links between LE-associated stromal states and pro-invasive signaling programs. Furthermore, we developed SpaPred, which demonstrated favorable performance in inferring the spatiotemporal heterogeneity of HCC at the spatial resolution. This model overcomes the limitations of existing algorithms in analyzing the composition of tissue spatial structures. Finally, integration of in silico trajectory-perturbation and pharmacogenomic drug-response analyses prioritized Oxaliplatin, Belinostat, and Temsirolimus as candidate compounds associated with LE-related transcriptional programs. These drug predictions are computational and require experimental validation. Collectively, SpaPred provides a hypothesis-generating framework for investigating spatial heterogeneity and candidate therapeutic vulnerabilities in HCC.

The current reliance on wet-lab experiments for evaluating the efficacy and adverse effects of natural products remains a major obstacle in new drug discovery. We propose KAN-PROSPECT, a novel deep learning framework that integrates transfer learning with Kolmogorov-Arnold Networks (KAN). This model can simultaneously predict the efficacy and adverse effects of natural products based solely on molecular SMILES representations, thereby addressing the limited generalizability of existing models in predicting these aspects for natural products. Beyond methodological advances, KAN-PROSPECT also contributes to more sustainable and resource-efficient drug discovery. Leveraging a cross-modal transfer learning strategy pretrained on approximately 3,800 drugs and fine-tuned on 400 natural products, KAN-PROSPECT consistently outperforms baseline models in dual-label prediction tasks. Notably, it demonstrates exceptional robustness in addressing data scarcity, excelling particularly in few-shot and zero-shot scenarios. Through the transfer learning strategy, the model partially alleviates the data scarcity issue commonly encountered in natural product prediction tasks. In addition, the incorporation of KAN layers enhances the ability to model complex nonlinear relationships between molecular structures and associated pharmacological or adverse reaction profiles, contributing to improved predictive performance. Furthermore, the framework demonstrates strong generalization ability, enabling high-accuracy predictions for the efficacy and adverse effects of entirely new natural products. KAN-PROSPECT was further applied to comprehensively predict natural products from the MEC and NPASS databases, with Icaritin from Epimedium used as a representative case study. Overall, KAN-PROSPECT is the first framework to unify transfer learning with the KAN architecture for dual-label prediction of natural products, showing great potential for large-scale bioactivity and toxicity prediction, new drug development, and drug repositioning of natural products.

The prediction of drug response would significantly improve the treatment of lung cancer. Tumor heterogeneity and complex signal transduction pathways lead to varied treatment effects among patients, but traditional computational approaches struggle to model the nonlinear, high-dimensional relationship between genes and drug responses. In order to develop a Generative Adversarial Network (GAN)-based model that can predict drug-induced gene expression profiles from lung cancer cell lines, we developed GRIP-Lung (Generative Model of Response to Drug-Induced Perturbation in Lung Cancer). By making use of biologically informed embeddings of cell line identity as well as drug treatment conditions, this model is able to gain a fairly good understanding of cell types and their transcriptional perturbations induced by different drugs. The GRIP-Lung model displayed reasonably good prediction ability in terms of predictive accuracy and showed high concordance between the predicted and experimental expression profiles. We not only predicted transcriptional changes induced by drug therapy but also used single-sample Gene Set Enrichment Analysis (ssGSEA) to classify post-treatment response states based on characteristic molecular biomarkers, offering a means for selecting effective drugs to target specific heterogeneity within lung tumors. The proposed GRIP-Lung framework faithfully reproduces drug-induced transcriptional perturbations in lung cell line models. By integrating biologically informed embeddings and adversarial learning, the model advances drug response prediction. This makes it a flexible computational tool for drug repositioning.

While the recent FDA-approved antiandrogen enzalutamide (Enz) might prolong the survival of castration-resistant prostate cancer (CRPC) patients by an additional 4.8 months, most patients eventually might still develop Enz resistance within 6–12 months. Although few genes have been linked to Enz resistance in prostate cancer (PCa), the detailed mechanism(s) are still underinvestigated. Here, we found that Enz might function by altering androgen receptor(AR)-mediated CDR1/circCDR1-AS/miR-1290/BMP4 signaling to modulate PCa stem cells (CSCs) to increase Enz resistance. Mechanistic analysis revealed that Enz/AR signaling can transcriptionally regulate CDR1 expression by reducing binding to androgen response elements (AREs) on the CDR1 5’ promoter to alter circCDR1-AS expression. Enz/AR/CDR1/circCDR1-AS signaling might then increase BMP4 expression by altering miR-1290 expression, which involves direct binding to the 3’ UTR of BMP4 mRNA. Preclinical studies using a CWR22Rv1 xenograft mouse model and integrative analysis of GEO cohort data further demonstrated that targeting this newly identified Enz/AR/CDR1/circCDR1-AS/miR-1290/BMP4 signaling pathway with miR-1290, circCDR1-AS-shRNA, or BMP4-shRNA may help develop novel therapies to combat Enz resistance at the later stage of CRPC.

Background: Metastasis, the spread of cancer cells from the primary tumor to distant organs, is the leading cause of mortality in cancer patients. This process often exhibits a preference for specific organs, a phenomenon known as tumor organotropism. This study focuses on the organotropism of breast cancer and analyzes its genomic alterations following metastasis to four organs (bone, brain, liver, and lung). The research aims to explore the intrinsic characteristics of primary breast cancer and the interactions between tumor cells and the tumor microenvironment (TME) within these target organs. Building upon this foundation, we developed a deep learning model to identify organ-specific metastatic genes, providing insights into the molecular mechanisms of metastasis. Methods: To investigate the mechanisms of organ-specific metastasis in breast cancer, we employed an integrative approach combining single-cell RNA sequencing, bulk RNA sequencing, ChIP-seq data, and deep learning techniques. Single-cell analysis provided detailed insights into cellular heterogeneity and microenvironment interactions at metastatic sites. Bulk RNA sequencing enabled the identification of gene expression patterns associated with metastatic propensity. A deep neural network (DNN) model was developed to analyze these complex datasets and identify key predictors of organ-specific metastasis. Results: Our integrative analysis revealed distinct gene expression profiles and cellular compositions in metastatic lesions across different organs. We have identified that, regardless of the target organ, breast cancer metastasis critically depends on specific biological signaling pathways, including the MAPK signaling pathway, metabolic pathways, the PI3K-Akt signaling pathway, and the positive regulation of cell adhesion. Single-cell sequencing highlighted unique interactions between tumor cells and the microenvironment, which varied significantly depending on the metastatic site. Fibroblasts play a critical role in facilitating the colonization of breast cancer cells in metastatic organs. The deep learning models effectively identified key molecular signatures and pathways associated with organ-specific metastasis, providing insights into the metastatic process. Conclusion: The study underscores the importance of the tumor microenvironment in influencing breast cancer metastasis to distant organs. We also established a comprehensive framework for understanding the mechanisms driving organotropism metastasis in breast cancer. Additionally, we identified key genes and signaling pathways associated with organ-specific metastasis, providing insights that may inform future studies on risk assessment and potential therapeutic targets for metastatic breast cancer.

Hypoxia, a common feature in many malignancies, is particularly prominent in cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC). Investigating the mechanisms underlying hypoxia is essential for understanding the heterogeneity of CESC and developing personalized therapeutic regimens. Firstly, the CESC-specific hypoxia gene sets shared between single-cell RNA sequencing (scRNA-seq) and bulk data were identified through Weighted Gene Correlation Network Analysis (WGCNA)and FindMarkers analyses. A CESC-specific hypoxia-related score (CSHRS) risk model was constructed using the least absolute shrinkage and selection operator (LASSO)and Cox regression analyses based on these genes. The prognostic differences were analyzed in terms of immune infiltration, mutations, and drug resistance. Finally, a nomogram model was constructed by integrating clinicopathological features to facilitate precision treatment for CESC. This study constructed a CSHRS risk model that divides patients into two groups, and this model can comprehensively evaluate the tumor microenvironment characteristics of CESC, provide accurate prognostic predictions, and offer rational treatment options for patients.

Hepatocellular carcinoma metastasis occurs mainly in the portal vein and lymph node metastasis, and is the main cause of patient death. However, the cellular origins and driving mechanisms of hepatocellular carcinoma metastasis have not been elucidated. In this study, we found that different tumor metastasis samples originated from different branches of the primary tumor subclone. A large number of sequence data confirmed the correlation between tumor metastasis index and metastasis and prognosis. Patients with a high index generally had a poor prognosis and abnormal metabolic function. Finally, we recommended a candidate therapy for each metastatic direction. This study explains the cellular origin and underlying mechanisms of hepatocellular carcinoma metastasis at the single-cell level and identifies drugs for its targeted therapy. It also provides a new research idea for the study of hepatocellular carcinoma metastasis and early identification of cancer metastasis from cellular resolution.