Esophageal squamous cell carcinoma (ESCC) remains one of the most prevalent malignancies worldwide and is associated with an extremely poor prognosis. Gastric cardia adenocarcinoma (GCA), arising at the junction between the esophagus and stomach, exhibits distinct histological characteristics and clinical outcomes compared with ESCC. Both cancers show exceptionally high incidence rates in the central-northern regions of China. Despite their clinical importance, the molecular mechanisms underlying these cancers, including genomic mutations, structural variations, and tumor microenvironmental changes, remain incompletely understood. Building upon our previous publication in Nature Communications [PMID: 34764264], in which we demonstrated that high ERBB2 amplification or expression in GCA is associated with improved prognosis—contrary to observations in many other cancer types—we further expanded our multi-omics investigation during the current reporting period. We collected and analyzed a large cohort comprising more than 100 tumor samples with matched transcriptomic, proteomic, and metabolomic data. Comprehensive analyses were performed to characterize mutation profiles, genomic structural variations, differential gene expression patterns, and immune cell compositions among ERBB2-High, ERBB2-Low, and ERBB2-Negative groups. Our findings revealed substantial molecular heterogeneity associated with different ERBB2 expression levels, suggesting distinct biological mechanisms and tumor progression pathways. In addition, we performed single-cell RNA sequencing (scRNA-seq) on 31 tumor samples to further dissect the cellular composition and tumor microenvironment at single-cell resolution. Cell populations were identified and annotated across all samples, and comparative analyses were conducted between ERBB2-High and ERBB2-Low groups. These analyses identified significant differences in immune cell infiltration patterns, epithelial cell states, and stromal cell interactions associated with ERBB2 expression status. Asides from the ESCC project, we currently have multiple projects focusing on studying the initiation and progression of different forms of cancers employing 10x Visium spatial transcriptome platform. Currently, data was generated from gastric cardia adenocarcinoma (n=24), cervical cancer (n=20), invasive mucinous adenocarcinoma (n=12) and esophagus squamous carcinoma with metastasis to the brain (n = 24). To analyze the spatial transcriptomic data, we routinely use machine learning based packages/softwares that are powered by the use of GPU. the analysis of spatial trancriptome data heavily relies on the correct annotation of relevant scRNA-seq data from our own projects and previous studies. To annotate and merge the scRNA-seq data from different studies, we usually use packages like Seurat or scverse (especially scvi-tools) for batch correction, integration and label transfer. For spatial transcriptome analysis, we rely on packages utilizing pyTorch framework to perform our analysis. Furthermore, we use tools such as Cell2location, Spotiphy, TopoVelo and scNiche to infer tumor microenvironment and cell lineages. Currently, we are learning and developing new tools to analyze our data. These tools generally utilize GPU resources for rapid model training. As the above mentioned projects are still ongiong, we are applying for extending the projects NAISS 2025/5-333, NAISS 2025/23-520 and NAISS 2025/6-228 for another year.