Method for detecting potential protein biomarker and drug target of gastric cancer

Abstract

According to the screening method for the potential protein biomarkers and the drug targets of the gastric cancer, Mendel randomization analysis in a proteome range is adopted, the genetic causal relationship between circulating plasma protein and the risk of the gastric cancer is evaluated, and finally the remarkably related protein is identified. According to the screening method of the potential protein biomarker and the drug target of the gastric cancer, provided by the invention, the potential association between circulating plasma protein and the gastric cancer is systematically revealed by integrating Mendel randomization, single-cell RNA sequencing analysis, space transcriptome analysis, virtual drug screening, molecular docking, molecular dynamics simulation and other methods.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of biological medicine, and particularly relates to a detection method for a gastric cancer potential protein biomarker and a drug target. BACKGROUND

[0002] Gastric cancer is the fifth most common cancer worldwide and the third leading cause of cancer death, causing more than 784,000 deaths each year. Due to the low rate of early diagnosis, most patients are diagnosed at an advanced stage. For these patients with advanced gastric cancer, targeted therapy has become a key treatment. At present, the main targeted therapies for gastric cancer include anti-human epidermal growth factor receptor 2 (HER2) therapy, anti-angiogenesis therapy and anti-programmed death 1 (PD-1) therapy. However, a major challenge of targeted therapy is the development of drug resistance, which limits the long-term efficacy of these treatments and highlights the need for continuous exploration of new therapeutic targets.

[0003] As an important component in circulating blood, plasma proteins are derived from both cell leakage and active secretion, participate in various key physiological and pathological processes, can reflect the overall state of the body, and can become potential drug targets. Previous studies have shown that many circulating proteins can be used as biomarkers for inflammation, infection and some systemic diseases. In the field of malignant tumors, many cross-sectional studies have attempted to reveal the complex relationship between proteins and tumor occurrence by comparing the differences in circulating protein levels between cancer patients and healthy controls. However, due to the inherent confounding bias and reverse causality of observational studies, the reliability of their conclusions is limited.

[0004] In recent years, a number of large-scale proteomic studies have identified protein quantitative trait loci (pQTL), making it possible to use two-sample Mendelian randomization (MR) methods to infer the causal effect of plasma proteins on GC susceptibility. Similar to randomized controlled trials, MR uses genetic variation as a tool variable, so it is less affected by confounding factors and reverse causality, and has been widely used to explore the causal relationship between, for example, peripheral metabolites or intestinal microbiota and GC. However, there is no report on the use of two-sample Mendelian randomization (MR) method to detect new protein biomarkers and drug targets for gastric cancer. SUMMARY

[0005] Technical problem: In order to solve the defects of the prior art, the application provides a detection method for a gastric cancer potential protein biomarker and a drug target.

[0006] Technical solution: The detection method for a gastric cancer potential protein biomarker and a drug target provided by the application comprises the following steps:

[0007] (1) Mendelian randomization analysis: using the genome-wide association study (GWAS) data of gastric cancer (GC) and cis-pQTL data, trans-pQTL data from public databases, performing two-sample Mendelian randomization (MR) analysis in the proteome range to obtain the causal relationship between plasma proteins and gastric cancer;

[0008] (2) PPI network construction and key target screening: performing protein-protein interaction (PPI) analysis on the 29 plasma proteins obtained in step (1) that have a causal relationship with gastric cancer (GC) to screen for biologically meaningful interaction relationships; visualizing the analysis results and calculating node parameters to screen for the top-ranking key targets;

[0009] (3) Verification of key targets: using single-cell RNA sequencing (scRNA-seq) analysis on high-quality cells screened, clustering and type annotating the cells, analyzing hub genes, and analyzing the expression distribution of hub genes in each cell type; using Kaplan-Meier clinical survival analysis to annotate the association between hub gene mRNA levels and gastric cancer patient prognosis.

[0010] In step (1), the cis-pQTL data and trans-pQTL data are obtained from the deCODE database (https: / / www.decode.com, derived from the study by Ferkingstad et al. (Ferkingstad et al., 2021), which used 4,907 aptamers to perform GWAS analysis on plasma protein levels in a cohort of 35,559 Icelanders); the genome-wide association study (GWAS) data of gastric cancer is obtained from the IEU GWAS database (https: / / gwas.mrcieu.ac.uk, dataset code ebi-a-GCST90435577, this gastric cancer GWAS includes 393,926 subjects, including 393,372 controls and 554 cases, covering a total of 35,642,134 SNPs, and the study population is of European descent).

[0011] In step (1), in the two-sample Mendelian randomization (MR) analysis, single nucleotide polymorphism (SNP) data as instrumental variables (IVs) are used, which are mixed plasma protein quantitative trait locus (pQTL) data, including cis-pQTL data and trans-pQTL data. The SNP data are screened by the following methods: a significance threshold of P < 5 x 10 -8 ; the F statistic of each SNP is calculated, and only SNP data with an F statistic greater than 10 are retained; the clumping algorithm is executed using the “TwoSampleMR” R package with parameters set as r 2 = 0.001, distance 10,000 kb; when SNP data in the exposure group are missing in the outcome data set, proxy variants with strong LD (r 2 > 0.8) of the original SNP are selected to replace them.

[0012] In step (1), the two-sample Mendelian randomization (MR) analysis includes:

[0013] (a) Statistical analysis: completed using R software (version 4.3.2);

[0014] (b) MR analysis: performed by the “TwoSampleMR” package (Hemani et al., 2018);

[0015] (c) Inverse-variance weighted (IVW) analysis: as the main method for evaluating the causal relationship between plasma proteins and gastric cancer, the effect is estimated by weighted regression of SNP-outcome correlation and SNP-exposure correlation, with the intercept assumed to be zero (Burgess et al., 2013);

[0016] (d) Wald ratio method analysis: when a protein has only a single SNP that meets the requirements, the Wald ratio method is used;

[0017] (e) False discovery rate (FDR) correction: to reduce false positives, the P value is subjected to false discovery rate (FDR) correction, and plasma proteins with a corrected P value less than 0.05 are considered significant positives;

[0018] (f) Sensitivity analysis: To ensure the validity and robustness of the results, sensitivity analysis was performed, including heterogeneity analysis and pleiotropy analysis. Pleiotropy analysis mainly assessed whether there was horizontal pleiotropy among multiple IVs. The presence of directional pleiotropy was determined by the intercept test of MR-Egger regression. When the intercept was significantly deviated from zero (Bowden et al., 2015), it indicated the presence of directional pleiotropy (Bowden et al., 2019). In the heterogeneity analysis, Cochran's Q value was used to measure the degree of difference among IVs. When the Q value was high, it indicated the presence of significant variability (Verbanck et al., 2018a).

[0019] In step (2), specifically: using the STRING database (https: / / cn.string-db.org / ) to perform Protein-Protein Interaction (PPI) analysis on plasma proteins with causal relationships with gastric cancer, setting the confidence score threshold to be ≥0.4 to screen for biologically meaningful interaction relationships; using Cytoscape software (version number 3.10.3, https: / / cytoscape.org / ) to visualize the analysis results and calculate node parameters, using the CytoHubba plugin in Cytoscape, combined with Edge Percolated Component (EPC) and Degree Centrality (Degree, number of node connections), to screen the top 10 key target points; constructing a Venn diagram to cross-screen the key target points to extract core target points.

[0020] In step (3), specifically: using the Seurat R package (Hao et al., 2024) for single-cell RNA sequencing (scRNA-seq) analysis, screening high-quality cells; clustering and dividing the cells; annotating the cell types; analyzing the expression distribution of related proteins in each cell type.

[0021] Preferably, the high-quality cells need to meet the following conditions: mitochondrial gene content is less than 20%, the number of expressed genes is more than 200, the total number of detected genes is between 200 and 6000, and at least appears in three cells;

[0022] Wherein, the method for clustering and dividing the cells is: using the Harmony algorithm for data integration to correct batch effect and improve clustering accuracy; the data is processed by log-normalization, and the top 2000 high-variation genes are screened out by using the FindVariableFeatures function; then principal component analysis (PCA) is performed for dimension reduction, and soft k-means clustering is implemented by the Harmony package; the cells are clustered and divided by the FindClusters function (resolution is set to 0.4);

[0023] The cell type annotation method is: determined according to classical marker genes, differential expression patterns and known cell characteristics.

[0024] Further comprising step (4), using gene ontology (GO) functional annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis to obtain the potential role of key genes in biological processes, molecular functions and / or cell components and / or the signal pathways involved by key genes; in step (4), the gene ontology (GO) functional annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis are specifically: using the online DAVID database (https: / / davidbioinformatics.nih.gov / ) to perform gene ontology (GO) functional annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis on the plasma proteins related to gastric cancer, to obtain the potential role of key genes in biological processes, molecular functions and cell components, and the signal pathways they are involved in; using the online data analysis and visualization platform (https: / / www.bioinformatics.com.cn) to make the results more intuitively presented.

[0025] The application further provides a gastric cancer potential protein biomarker and a drug target, including BMP4, GRP, LGR4, MGP, TGFB3, ERBB3 or ETS2.

[0026] The application further provides an application of BMP4 as a gastric cancer potential protein biomarker or a drug target.

[0027] The application further provides an application of GRP as a gastric cancer potential protein biomarker or a drug target.

[0028] The application further provides an application of LGR4 as a gastric cancer potential protein biomarker or a drug target.

[0029] The application further provides an application of MGP as a gastric cancer potential protein biomarker or a drug target.

[0030] The application also provides application of TGFB3 as a potential protein biomarker or drug target of gastric cancer.

[0031] The application also provides application of ERBB3 as a potential protein biomarker or drug target of gastric cancer.

[0032] The application also provides application of ETS2 as a potential protein biomarker or drug target of gastric cancer.

[0033] The application also provides screening of the ZINC database for compounds by using the hub gene screened by the above method and virtual screening of drugs.

[0034] Beneficial effects: The screening method for the potential protein biomarker and drug target of gastric cancer provided by the application systematically reveals the potential correlation between circulating plasma proteins and gastric cancer by integrating Mendelian randomization, single-cell RNA sequencing analysis, spatial transcriptome analysis, virtual screening of drugs, molecular docking and molecular dynamics simulation, etc. The research results show that eight key genes, i.e., ERBB3, LGR4, BMP4, CD248, MGP, TGFB3, GRP and ETS2, may be important hubs affecting the occurrence and development of gastric cancer. These findings provide a new theoretical basis for in-depth understanding of the molecular mechanism of circulating plasma proteins affecting the progression of gastric cancer. In addition, these key genes have good potential application value in early warning, prognosis evaluation and targeted treatment of gastric cancer. Future research will further explore the functional mechanism of these genes and their potential application in the clinic.

[0035] The screening method for the potential protein biomarker and drug target of gastric cancer provided by the application adopts Mendelian randomization (MR) analysis in the proteome range, systematically evaluates the genetic causal relationship between 4907 circulating plasma proteins and the risk of gastric cancer (GC), and finally identifies 8 significantly related proteins. These findings not only provide new insights into the pathogenesis of gastric cancer, but also provide potential candidate targets for targeted treatment and biomarker development of gastric cancer.

[0036] In this study, we integrated cis- and trans-pQTL data to perform MR analysis on a proteome-wide scale to identify plasma proteins associated with GC. Subsequently, we performed sensitivity analysis to validate the robustness of the preliminary results. We also analyzed the causal relationship between the expression of the corresponding protein-coding genes and GC, and explored their expression patterns in single-cell types. We performed functional enrichment analysis, clinical survival analysis, and protein-protein interaction network construction to preliminarily explore the functions and related pathways of these targets. Finally, based on the screened hub genes, we performed virtual screening on the compound database to prioritize the identification of druggable candidate compounds, and verified the stable binding between the hub genes and the compounds through molecular docking and molecular dynamics simulation, laying a foundation for the development of clinical drugs.

[0037] ERBB3 and ETS2 identified in the present application exhibit a protective effect on gastric cancer, which is consistent with the bioinformatics analysis results of the present application. ERBB3 is a tyrosine kinase receptor (Kiavue et al., 2020), which has been shown to play an anticancer role in various tumors by inhibiting tumor cell proliferation and promoting cell differentiation (Sergina et al., 2007). ETS2 is a transcription factor, and literature indicates that it has tumor suppressor function in certain tumor backgrounds (Kabbout et al., 2013), which may be achieved by participating in the regulation of cell cycle and apoptosis-related genes (AFry and Inoue, 2018). On the contrary, BMP4, CD248, GRP, LGR4, MGP and TGFB3 are shown to be pro-cancer related factors in the present application, suggesting that they may play an important role in the development of gastric cancer by promoting cell proliferation, metastasis or immune escape, etc. BMP4 has been confirmed to activate the TGF-β signaling pathway in multiple cancers (Deng et al., 2020), promote epithelial-mesenchymal transition and cell migration (F.Xu et al., 2021). CD248 is a tumor-associated fibroblast surface marker (Wu et al., 2022), which has a key function in tumor angiogenesis and matrix remodeling (DiBenedetto et al., 2019). GRP is a neuropeptide growth factor (Li et al., 2013), which can stimulate cancer cell proliferation and metastasis in various cancers (Ischia et al., 2014). LGR4 and MGP are related to tumor progression, angiogenesis and immune microenvironment regulation in multiple solid tumors (Yue et al., 2018) (Li, 2024). TGFB3 is a member of the TGF-β family (Zelisko et al., 2024), which is widely involved in cell growth, apoptosis and immune escape (Qin et al., 2016), and its high expression is usually associated with poor prognosis (Alkhathami et al., 2025).

[0038] In the present application, single-cell RNA sequencing results further verify the expression characteristics of the above key proteins in different gastric cancer cell subpopulations, highlighting their functional heterogeneity in different cell types. For example, ERBB3 is mainly expressed in epithelial cells, which may exert an anticancer effect by maintaining epithelial homeostasis (T. Xu et al., 2021), while the high expression of CD248 in fibroblasts may promote tumor-associated stromal formation (Fujii et al., 2015). BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 . Study design flowchart. Among them, MR: Mendelian randomization; IV: instrumental variable; pQTL: protein quantitative trait loci; GWAS: genome-wide association study; PPI: protein–protein interaction; OS: overall survival; KEGG: Kyoto Encyclopedia of Genes and Genomes; GO: Gene Ontology; EPC: Edge Percolated Component.

[0040] Figure 2A. Forest plot showing the results of Mendelian randomization (MR) analysis of 29 circulating plasma proteins significantly causally associated with gastric cancer risk. The effect size of each protein is represented by odds ratio (OR), and the horizontal line is the 95% confidence interval. The analysis method is IVW (Inverse Variance Weighted), and the P value is a two-sided test result. All proteins shown pass the FDR correction (P < 0.05). B. Circular heat map showing the P value distribution of candidate proteins under different MR methods (IVW, MREgger, weighted median, weighted mode, simple mode). Each color block represents the P value size under a method, and the redder the color, the stronger the statistical significance (the smaller the P value). It can be seen from the figure that the 29 circulating plasma proteins significantly associated with gastric cancer risk have stable significance in multiple methods, indicating that the results are robust. C. Volcano plot showing the results of MR analysis of deCODE circulating plasma proteins and gastric cancer risk. The X-axis is the effect value (Beta), and the Y-axis is -log10(P value). Each point represents a protein. The color is stratified according to significance: gray represents not significant after FDR correction (P > 0.05), blue-green-yellow represents different significance levels (P < 0.05), and the redder the color of the related protein, the stronger the significance, indicating that the genetic causal relationship is highly reliable. D. Manhattan plot showing the MR significance distribution of all proteins in the whole genome range. The X-axis represents the chromosome number, and the Y-axis is -log10(P value). Each point represents a protein signal. Red dots represent proteins that are still significant after FDR correction, and the position of the related protein in the figure is significant, indicating its potential causal role in gastric cancer susceptibility.

[0041] Figure 3 A. GO enrichment analysis results. B. KEGG enrichment analysis results. C. PPI network of 29 circulating plasma proteins significantly associated with gastric cancer. D. Top 10 degree centrality (Degree) analysis results. E. Top 10 edge percolation component (EPC) analysis results. F. Venn diagram of gastric cancer core target points.

[0042] Figure 4 A. Violin plot of the number of genes per cell (nFeature_RNA), UMI count per cell (nCount_RNA), and mitochondrial read fraction (percent.mt) detected in each cell in the 9 primary gastric tumor samples before quality filtering. B. Application of quality control thresholds (200 ≤ nFeature_RNA ≤ 6,000; percent.mt ≤ 20%). C. PC standard deviation inflection point plot of pc1-20. D. PCs before and after coordination batch heat map. E. System clustering dendrogram based on PC corrected by coordination. F. UMAP depicts 16 discrete cell populations.

[0043] Figure 5 A. Marker genes for cell identification. B. Eight cell clusters labeled and annotated by marker genes. C and D show expression of hub genes in each cell type.

[0044] Figure 6 A. Network graph and accompanying heatmap illustrating the number of cell-cell interactions between eight identified cell types. B. Network graph and accompanying heatmap illustrating the strength of cell-cell interactions between eight identified cell types. C. Quantitative description of the number and strength of cell-cell communication events between each identified cell type. D. Trajectory plot illustrating the pseudo-temporal arrangement of 16 identified cell populations. E. Plot depicting the time course of cell development along the pseudo-temporal axis. F. Independent visualization of cell type composition at each discrete pseudo-temporal stage. G. Trajectory plot depicting differentiation stages in pseudo-temporal succession. H. Pseudo-temporal trajectory plot describing the differentiation dynamics of eight identified cell types.

[0045] Figure 7 A. Spatial distribution of eight hub genes in gastric cancer tissue section GSM7990473. Areas showing stronger red color correspond to higher expression levels of the plotted genes. B. Spatial distribution of eight hub genes in gastric cancer tissue section GSM7990474. Deeper red color indicates increased local gene expression.

[0046] Figure 8 Overall survival (OS) analysis of GC patients using the Kaplan-Meier Plotter to assess the prognostic value of the following genes: A. BMP4, B. CD248, C. GRP, D. LGR4, E. MGP, F. TGFB3, G. ERBB3, and H. ETS2. Figure 9 A. 2D chemical structure of Proflavine hemisulfate. B. 2D chemical structure of Hydroxychloroquine sulfate. C. 2D chemical structure of Rizatriptan benzoate. D. 2D chemical structure of L-Histidine. E. 2D chemical structure of Retigabine.

[0047] Figure 10A. Schematic representation of the molecular docking of TGFB3 with proflavine hemisulfate. B. Schematic representation of the molecular docking of TGFB3 with hydroxychloroquine sulfate. C. Schematic representation of the molecular docking of TGFB3 with rizatriptan benzoate. D. Schematic representation of the molecular docking of TGFB3 with L-histidine. E. Schematic representation of the molecular docking of TGFB3 with retigabine. Left panel shows the binding pocket, right panel is a 2D interaction diagram. F. Heatmap displaying the binding energies (kcal / mol) between TGFB3 and proflavine hemisulfate, hydroxychloroquine sulfate, rizatriptan benzoate, L-histidine and retigabine, obtained by molecular docking analysis.

[0048] Figure 11 A. Root mean square deviation (RMSD) of the complex over the course of the 100 ns simulation. B. Total radius of gyration (Rg) and its components along the X, Y, Z axes (Rg_x / N, Rg_y / N, Rg_z / N) as a function of simulation time. C. Solvent accessible surface area (SASA) as a function of time. D. Two-dimensional Gibbs free energy landscape based on principal component analysis (PCA) (PC1-PC2), with color gradient representing the magnitude of free energy (kJ / mol). E. Atom motion covariance matrix heatmap, with color intensity reflecting the positive or negative correlation strength between residues. F. Three-dimensional Gibbs free energy surface and its bottom heatmap projection, which intuitively displays the energy distribution of the complex along the PC1 / PC2 directions. DETAILED DESCRIPTION

[0049] The present application is further illustrated below.

[0050] Example 1

[0051] I. Experimental Methods

[0052] 1. Study dataset and genetic instrumental variable selection

[0053] The genome-wide association study (GWAS) data of gastric cancer was obtained from the IEU GWAS database (https: / / gwas.mrcieu.ac.uk) with dataset code ebi-a-GCST90435577. The gastric cancer GWAS included 393,926 subjects, including 393,372 controls and 554 cases, covering a total of 35,642,134 SNPs, and the study population was of European descent.

[0054] Plasma proteomics data were obtained from the deCODE database (https: / / www.decode.com) and derived from a study by Ferkingstad et al. (Ferkingstad et al., 2021) that performed a GWAS analysis of plasma protein levels in a cohort of 35,559 Icelanders using 4,907 aptamers. Given that the original studies of both exposure (plasma proteome) and outcome (gastric cancer) data in this MR study were conducted in different countries, it can be assumed that there is no sample overlap between the study populations. This study was conducted in accordance with the Declaration of Helsinki (revised in 2013).

[0055] The selection of IVs followed three widely accepted guidelines (Davies et al., 2018): first, IVs must be significantly associated with the exposure factor (in this study, plasma protein expression); second, IVs should not be associated with potential confounders; third, IVs should only affect the outcome through known risk factors and not via other pathways. In MR studies, single nucleotide polymorphisms (SNPs) are commonly used as IVs. SNPs refer to single base differences in DNA sequences between individuals and have the following advantages: they are stable throughout an individual's life, randomly assigned during gamete formation, and easy to detect (Hwang et al., 2019). These characteristics make SNPs ideal IVs.

[0056] The SNPs used in this study were mixed pQTLs, including both cis-pQTLs and trans-pQTLs. To select SNPs that meet the above criteria, we set a significance threshold of P < 5 x 10 -8 To address the problem of linkage disequilibrium (LD) and ensure the independence of SNPs, we used the clumping algorithm in the "TwoSampleMR" R package with parameters set to r 2 To prevent weak instrument bias, we calculated the F-statistic for each SNP and only kept SNPs with F-statistics greater than 10. In addition, when a SNP in the exposure group was missing in the outcome data set, we selected a proxy variant with strong LD (r 2

[0057] We performed MR analysis on 4,907 plasma proteins to explore their association with gastric cancer. Statistical analysis was performed using R software (version 4.3.2), and MR analysis was performed using the "TwoSampleMR" package (Hemani et al., 2018).

[0058] ​The inverse-variance weighted (IVW) method was used as the primary method to assess the causal relationship between plasma proteins and gastric cancer. This method estimates the effect by weighted regression of SNP-outcome associations and SNP-exposure associations, with an intercept assumption of zero (Burgess et al., 2013). When a protein has only a single eligible SNP, the Wald ratio method was used.

[0059] To reduce false positives, we performed false discovery rate (FDR) correction on P values, and considered plasma proteins with a corrected P value less than 0.05 as significant positives.

[0060] To ensure the validity and robustness of the results, we also performed several sensitivity analyses, including heterogeneity analysis and pleiotropy analysis. Pleiotropy analysis mainly assesses whether there is horizontal pleiotropy among multiple IVs, and the presence of directional pleiotropy is determined by the intercept test of MR-Egger regression—when the intercept significantly deviates from zero (Bowden et al., 2015), indicating the presence of directional pleiotropy (Bowden et al., 2019). In the heterogeneity analysis, Cochran’s Q value is used to measure the degree of difference among IVs, and when the Q value is high, it indicates the presence of significant variability (Verbanck et al., 2018a).

[0061] 3. Functional and pathway enrichment analysis

[0062] We used the online DAVID database (https: / / davidbioinformatics.nih.gov / ) to perform gene ontology (GO) functional annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis on plasma proteins with a causal relationship with gastric cancer (GC). These analyses revealed the potential roles of key genes in biological processes, molecular functions, and cellular components, as well as the signaling pathways they participate in. To make the results more intuitive, we used the online data analysis and visualization platform (https: / / www.bioinformatics.com.cn), last accessed on May 20, 2025. This platform provides convenient data display tools, further enhancing the interpretability and visualization of the results.

[0063] 4. Screening of gastric cancer core target points and construction of PPI network

[0064] We performed protein-protein interaction (PPI) analysis on plasma proteins with causal relationships with gastric cancer (GC) using the STRING database (https: / / cn.string-db.org / ), setting the confidence score threshold to ≥0.4 to screen for biologically meaningful interaction relationships. Subsequently, we visualized the analysis results using Cytoscape3.10.3 software (https: / / cytoscape.org / ), calculated node parameters, and used the CytoHubba plug-in in Cytoscape3.10.3 to screen the top 10 key target points based on Edge Percolated Component (EPC) and Degree centrality (Degree, number of node connections).

[0065] Further, we cross-screened the key target points to extract core target points (hub genes) by constructing a Venn diagram.

[0066] 5. Single-cell RNA sequencing analysis

[0067] We performed single-cell RNA sequencing (scRNA-seq) analysis using the Seurat R package (Hao et al., 2024). The scRNA-seq data for gastric cancer patients and control individuals were obtained from the study by Jiang H et al. (Jiang et al., 2022), which constructed a gastric cancer single-cell cohort. The data used has been publicly released under GSE163558 in the GEO database. The dataset contains 10 samples, and we selected 9 gastric cancer samples (including distant metastasis) from them. We used strict quality control standards to ensure the reliability of the data.

[0068] The retained cells must meet the following conditions: mitochondrial gene content is less than 20%, the number of expressed genes is more than 200, the total number of detected genes is between 200 and 6000, and at least appears in three cells. Finally, the high-quality cells screened were used for subsequent analysis.

[0069] To correct batch effects and improve clustering accuracy, we integrated the data using the Harmony algorithm. The data were log-normalized and the top 2000 variable genes were selected using the FindVariableFeatures function. Subsequently, principal component analysis (PCA) was performed for dimensionality reduction, and soft k-means clustering was implemented using the Harmony package. The cells were clustered by the FindClusters function with a resolution of 0.4. Cell type annotation was determined based on classical marker genes, differential expression patterns, and known cell characteristics.

[0070] In addition, to explore the potential cellular mechanisms between the hub genes related to gastric cancer (GC) and the progression of gastric cancer (GC), we further analyzed the expression distribution of hub genes in each cell type.

[0071] 6. Cell-cell communication analysis

[0072] Ligand-receptor interactions between gastric cancer cell populations were analyzed using the CellChat R package (https: / / github.com / sqjin / CellChat) with default parameters. Single-cell transcriptomes were used as input, and by integrating these expression data with curated ligand-receptor-cofactor interactions from the CellChatDB repository, the cell-cell communication network was reconstructed.

[0073] 7. Pseudo-time series analysis

[0074] Monocle2 was used to perform pseudotime trajectory analysis on the normalized gene expression matrix. Highly variable genes were selected to reconstruct developmental trajectories, enabling the inference of dynamic changes during cell progression. This approach helped elucidate the lineage relationships and temporal transitions between gastric cancer cell populations.

[0075] The gastric cancer spatial transcriptome (ST-seq) dataset was retrieved from GSE251950. To map the spatial distribution of single-cell clusters within gastric tumors, we applied the CARD R package for spatial deconvolution of transcriptome profiles. In our workflow, a CARD object was first generated by the createCARDObject function, followed by the execution of CARD_deconvolution with default settings.

[0076] 9. Kaplan-Meier plot database analysis

[0077] This study used the Kaplan-Meierplotter database (www.kmplot.com) to explore the association between the mRNA levels of each core gene and the prognostic outcomes of gastric cancer patients. Specifically, patient samples were divided into high-expression and low-expression groups based on the median expression level of each gene to assess its prognostic significance. The platform automatically calculates the hazard ratio (HR), its 95% confidence interval (CI), and the log-rank test p-value, thus facilitating survival analysis (Lánczky et al., 2016).

[0078] 10. Virtual drug screening and molecular docking

[0079] The small molecule candidate compounds we used were derived from the ZINC database, and the 3D structures of the receptors were obtained from the UniProt database. ZINC provides commercially available compounds with good drug-likeness and 3D structural information. The candidate compounds were first processed through... Preprocessing was performed using the LigPrep module in Maestro 12.8 software, including adding hydrogen atoms, adjusting the ionization state at pH 7.0 ± 2.0, and energy optimization based on the OPLS2005 force field to obtain stable and reasonable three-dimensional structures (Zheng et al., 2024). The target protein was a gastric cancer (GC)-related protein (hubgenes), whose UniProt entries were used to determine protein information, and the corresponding three-dimensional structures (PDB format) were downloaded from the RCSBPDB database. The protein structures were then preprocessed using AutoDockTools, including removing water molecules, adding polar hydrogen atoms, calculating Gasteiger charges, and saving in PDBQT format; the compound structures were also converted to PDBQT format in the same way to meet the input requirements of AutoDockVina. After completing the preprocessing of ligands and proteins, the following steps were first performed using... The VirtualDrugScreening module in Maestro 12.8 (https: / / www.schrodinger.com, accessed May 16, 2025) performs initial screening of the pretreated ZINC compound library, removing compounds with unstable structures or low pharmacological potential. Subsequently, the selected candidate compounds undergo molecular docking analysis using AutoDockVina (Wang, 2025). During docking, a gridbox is constructed to cover the active pocket region of the target protein, with a size set to [size missing]. The docking score (in kcal / mol) output by AutoDock Vina was used to evaluate the binding ability between the compound and the target protein, and the lower the score, the stronger the binding force and the higher the priority. The docking scores of all compounds were ranked from high to low into a heatmap.

[0080] 8. Molecular dynamics simulation

[0081] Molecular dynamics simulation was performed using GROMACS version 2022.3 software (Abraham et al., 2015) (Van Der Spoel et al., 2005). The preprocessing of small molecule ligands was completed by AmberTools22, using GAFF force field parameterization; at the same time, Gaussian16W was used to hydrogenate the molecules and calculate the RESP charge. The simulation conditions were set to constant temperature (300 K) and standard atmospheric pressure (1 bar). The protein part was modeled using the Amber99SB-ILDN force field, the solvent environment was the TIP3P water model, and an appropriate amount of Na + ions were added to neutralize the system charge. The entire simulation system was first subjected to energy minimization by the steepest descent method, and then subjected to equilibrium simulation under constant temperature and volume (NVT) and constant temperature and pressure (NPT) conditions, respectively, with 100,000 steps for each stage, a coupling time constant of 0.1 ps, and a total duration of 100 ps. Subsequently, the system entered the formal free molecular dynamics simulation stage, running a total of 5,000,000 steps, with a time step of 2 fs, and a total simulation time of 100 ns. After the simulation was completed, the trajectory file was analyzed using the built-in tools of GROMACS, calculating the root mean square deviation (RMSD), solvent accessible surface area (SAS), and protein radius of gyration (Rg) of the amino acids in each trajectory, and combining the MM / GBSA method to calculate the free energy and generate a free energy landscape map, which was used to evaluate the binding stability and energy change trend of the complex (Shen et al., 2025).

[0082] II. Experimental results:

[0083] 1. Mendelian randomization analysis of gastric cancer-related circulating plasma proteins

[0084] 1.1 This study systematically evaluated the genetic causal relationship between 4907 circulating plasma proteins and gastric cancer (GC) risk using two-sample Mendelian randomization approach. By integrating deCODE proteomic data with the summary statistics of GC from IEU GWAS (ebi-a-GCST90435577), proteins with valid instrumental variables were finally included for MR inference. A total of 29 proteins were identified to be significantly causally related to GC risk (FDR corrected P<0.05), such as Figure 2 A, which includes multiple key molecules with known or potential functions in tumorigenesis. Significant results include: protective factors (OR<1): ERBB3 (OR=0.270, 95% CI: 0.121-0.603, P=0.001), LGR4 (OR=0.436, P=0.001), PCDH10, FSTL3, NDST1, etc. also showed significant protective effects. Risk factors (OR>1): TGFB3 (OR=8.083, P<0.001), CD248 (OR=2.074, P=0.005), BMP4, MGP, GRP, ETS2, etc. also significantly increased the risk of GC. As shown in Figure 2 B, the P-value distribution of the main candidate proteins under different MR methods (IVW, MR Egger, weighted median, weighted mode, simple mode) suggests that multiple proteins have robust statistical significance under multiple methods. The consistency of ERBB3, LGR4, TGFB3, CD248, BMP4, MGP, GRP, ETS2, etc. in different MR algorithms further supports their credibility as potential causal markers. As shown in Figure 2 C, the effect direction and significance of the candidate proteins, TGFB3, SPINK2, BIN1, LGR4, ERBB3, etc. are located at the top of the image due to their large effect size and significant P-value. As shown in Figure 2 .D, the proteins significantly associated with gastric cancer risk are shown by chromosome, further emphasizing the distribution characteristics and statistical significance of these proteins at the whole genome level. Overall, these results provide solid genetic evidence that multiple circulating proteins may play an important carcinogenic or protective role in the pathogenesis of gastric cancer, providing an important reference for subsequent functional verification and targeted therapy development.

[0085] 1.1. Functional annotation and pathway enrichment analysis

[0086] To further analyze the potential biological mechanisms of the 29 circulating proteins that are significantly causally related to gastric cancer risk, we conducted functional enrichment analysis based on GO and KEGG databases.

[0087] As shown in Figure 3A In terms of biological processes (BP), the enriched terms were mainly focused on ERBB3 signaling pathway, epithelial cell proliferation and differentiation, epidermis development regulation, trachea and skull system morphogenesis, and positive / negative regulation of protease activity, suggesting that these proteins might be involved in key biological processes such as epithelial tissue homeostasis and tissue remodeling. In terms of cellular components (CC), these proteins were mainly located in high-density lipoprotein particle, plasma lipoprotein particle, protein-lipid complex, endoplasmic reticulum lumen, proteasome regulatory subunit, and extrasynaptic vesicle membrane component, indicating that they might play a role in intracellular membrane system and secretory vesicle pathway. Molecular function (MF) analysis further revealed that these proteins had significant functional preference in protease inhibitor activity, endopeptidase regulation and inhibition activity, neuropeptide activity, and phosphatase function (such as PIP5K), suggesting their involvement in extracellular matrix regulation and signal transduction processes.

[0088] KEGG pathway analysis as Figure 3 B showed that these proteins were enriched in multiple cancer-related classical signaling pathways, including ErbB, MAPK, TGF-β, EGFR tyrosine kinase inhibitor resistance pathway, and colorectal cancer, gastric cancer, kidney cancer, and other tumor signaling networks. In addition, it was also involved in stem cell pluripotency maintenance (FoxO, Hippo), viral infection (such as HTLV-1, hepatitis B), metabolic pathway (nitrogen metabolism), and cell aging pathway. The above results collectively suggest that the identified proteins may play an important role in the development of gastric cancer by regulating epithelial development, lipid metabolism, protease activity, and various cancer-related signaling pathways.

[0089] 1.2 Gastric cancer core target screening and PPI network construction

[0090] Using the STRING database, we constructed a protein-protein interaction (PPI) network composed of 29 genes with significant causal relationships with gastric cancer (Fig. Figure 3 C). Subsequently, the topological properties of network nodes were analyzed in Cytoscape software, focusing on key indicators such as node degree (Degree) and edge percolation component (EPC). To systematically mine the core regulatory nodes in the network, we sorted all nodes according to their degree values, and used the color depth of the nodes to intuitively reflect their connectivity - the deeper the color, the more connected and active it is in the network. We called the cytoHubba plug-in (version 0.1) in the Cytoscape platform, based on Degree ( Figure 3 D), EPC ( Figure 3E) Two indices were used to filter the network nodes and each extracted the top 10 important nodes. By combining the screening results of the two indices, we finally confirmed 8 hub genes (BMP4, CD248, GRP, LGR4, MGP, TGFB3, ERBB3 and ETS2) Figure 3 F).

[0091] 1.3 Cell type-specific expression in gastric tumor tissues

[0092] We summarized the quality control indicators of nine primary gastric tumor samples in dataset GSE163558 by violin plots before and after filtering ( Figure 4 A-B). In the unfiltered data, the number of genes detected in each cell (nFeature_RNA) ranged from <200 to >12000 total UMI counts (nCount_RNA) reached 300000 or more, and the proportion of mitochondrial reads (percent.mt) was close to 95%, indicating a large proportion of low-quality or apoptotic cells. To address this issue, we applied strict filtering criteria to retain only cells that met all of the following conditions: 200≤n feature_RNA≤6000; nCount_RNA within the sample-specific distribution range. percent.mt≤20%; expression detected in at least three cells. Cells exhibiting low complexity or high mitochondrial content were excluded. After quality control, 20491 high-quality single cells remained. After filtering, nFeature_RNA values were concentrated between 200 and 6000, nCount_RNA values returned to a reasonable range, and percent.mt values consistently fell below 20%.

[0093] To optimize dimensionality reduction, we first generated a plot of the standard deviation of principal components (PCs) for PC10, which identified PC10 as the inflection point beyond which the explained variance would reach a stable level ( Figure 4 C). Therefore, the top ten PCs were retained for downstream analysis. Subsequently, batch effects were mitigated by Harmony integration, which was confirmed by the consistency of PC loadings before and after correction ( Figure 4 D), indicating that sample-specific differences were removed while retaining biologically relevant heterogeneity. Systematic clustering based on these 10 harmonized and corrected PCs produced a stable dendrogram with well-separated branches within a resolution range of 0.2-1.2 ( Figure 4 E). Finally, uniform manifold approximation and projection (UMAP) of the integrated and simplified dataset depicted 16 discrete transcriptional clusters ( Figure 4 F).

[0094] After dimensionality reduction clustering, cell subpopulations were identified and T cells, neutrophils, epithelial cells, macrophages, B cells, fibroblasts, proliferative cells, and endothelial cells were classified Figure 5 A). The identification of cell subpopulations was based on previously validated cell marker gene expression Figure 5 B)

[29] , including: T cells (CD2 and CD3D), neutrophils (S100A12 and HCAR3), epithelial cells (EPCAM and KRT19), macrophages (CD68 and CD163), B cells (CD79A, MS4A1, and IGHG1), fibroblasts (DCN and COL1A1), proliferative cells (MKI67 and TOP2A), and endothelial cells (VWF and PECAM1). In gastric tumor tissues, all 8 hub genes were expressed to varying degrees, Figure 5 C and Figure 5 D show the single-cell expression of these 8 hub genes in each cell cluster. Among them, ERBB3 was significantly highly expressed in epithelial cells; LGR4 was mainly moderately expressed in epithelial cells, and a small amount of expression was also detected in fibroblasts; BMP4 was expressed in a small amount in epithelial cells; CD248 was mainly significantly highly expressed in fibroblasts; MGP was significantly highly expressed in both fibroblasts and endothelial cells; TGFB3 was expressed in a small amount in fibroblasts; ETS2 was significantly highly expressed in neutrophils, epithelial cells, macrophages, and endothelial cells, and a small amount of expression was also detected in fibroblasts; GRP was expressed in a very small amount in fibroblasts.

[0095] 1.4 Cell-cell communication and pseudo-time trajectory analysis

[0096] Extensive evidence suggests that in the tumor microenvironment, malignant cells interact extensively with non-malignant stromal populations, most notably cancer-associated fibroblasts. Previous reports have demonstrated that stromal components, particularly inflammatory cafs, have profound effects on tumorigenesis and progression

[40] . Therefore, we reconstructed a cell-to-cell communication network to predict cell-cell signaling. Our analysis revealed multiple high-frequency, high-intensity interactions between macrophages and epithelial cells, fibroblasts and epithelial cells, macrophages and T cells, and fibroblasts and endothelial cells Figure 6 A–6C).

[0097] Pseudo-time trajectory inference (also known as cell state trajectory analysis) recapitulates the cell differentiation process by modeling the dynamic changes in the expression of temporally regulated genes in single-cell gene expression profiles. In this study, we constructed differentiation trajectories for the malignant cell population and identified 16 discrete cell clusters Figure 6 D). As Figure 6As shown in Fig. 6G and 6H, these eight cell types sequentially experienced five differentiation states; Figure 6 E and 6F further revealed the time sequence of subtype conversion, with cells migrating along the trajectory from deep to light pseudo-color. Specifically, T cells were mainly enriched in the earliest pseudo-time stage (state 1), at the root of the trajectory; as pseudo-time advanced, some T cells migrated along branch 4 to the intermediate state, but most still clustered near the starting point. Neutrophils were also mainly distributed in the early stage (state 1-2) at the base of branch 4. Fibroblasts extended along branch 2, which originated from the root, across the boundary between the early and intermediate stages (state 2). B cells were mainly located in the central intermediate stage (state 3) of branch 2. Cell populations in the “proliferation” state were enriched in the high-density region of the intermediate stage (state 3-4), representing a transition state that was about to differentiate along multiple terminal lineages. Endothelial cells were enriched in the middle-late stage (state 4), concentrated upstream of branch 3; epithelial cells were distributed in the late pseudo-time (state 4-5), migrating along branch 3 to the terminal position. Finally, macrophages almost only existed in the terminal pseudo-time stage (state 5), clustered at the top of branch 1.

[0098] 1.5 Spatial transcriptomic analysis of hub gene expression in gastric cancer tissues

[0099] To delineate the in situ expression pattern of the eight hub genes in gastric cancer tissues, we visualized their transcript abundance in two spatial transcriptomic sections (GSM7990473 and GSM7990474; Figure 7 A-B) In both specimens, TGFB3, BMP4, CD248, GRP, LGR4, MGP, ERBB3, and ETS2 signals were highly overlapped, mainly concentrated at the tumor-stroma interface. In section GSM7990473 Figure 7 A) CD248 and MGP outlined a high-expression band in the center of the section, flanked by intense TGFB3 and LGR4 signals, respectively; ERBB3 and ETS2 were relatively enriched in epithelial-like structures, while BMP4 and GRP were distributed throughout the section, but peaked in the above hotspots. In section GSM7990474 Figure 7 B) The co-enrichment of GRP and ERBB3 defined multiple discrete high-expression regions; TGFB3 and MGP formed clustered high-signal foci at the tissue edge; BMP4 and ETS2 overlapped multiple times in similar regions; CD248 and LGR4 reached expression peaks at adjacent positions. The consistent spatial structural features in the two sections indicated that the eight hub genes exhibited a common, locally concentrated transcriptional pattern in gastric cancer tissues, suggesting that these key regions are worth further functional verification.

[0100] 1.6 Survival analysis of hub genes in gastric cancer

[0101] Kaplan-Meier Plotter is a convenient online analysis tool for survival analysis based on gene expression levels to assess the prognostic relevance of key genes. In this study, gastric cancer patient samples were divided into high and low expression groups according to the median value of mRNA expression of each hub gene.

[0102] The results showed that high expression of BMP4 (HR = 1.39, 95% CI: 1.17-1.66, log-rank P = 0.00023), CD248 (HR = 1.48, 95% CI: 1.23-1.78, P = 3.4 x 10 -5 ), GRP (HR = 1.47, 95% CI: 1.22-1.76, P = 4.1 x 10 -5 ), LGR4 (HR = 1.46, 95% CI: 1.11-1.91, P = 0.0064), MGP (HR = 1.49, 95% CI: 1.24-1.80, P = 1.7 x 10 -5 ), and TGFB3 (HR = 1.71, 95% CI: 1.42-2.07, P = 1.6 x 10 -8 ) were significantly associated with worse overall survival (OS); in contrast, high expression of ERBB3 (HR = 0.69, 95% CI: 0.56-0.86, P = 0.00084) and ETS2 (HR = 0.60, 95% CI: 0.50-0.73, P = 7.8 x 10 -8 ) predicted better OS( Figure 10 ). These results suggest that the six genes mentioned above can be potential predictive biomarkers for poor prognosis in gastric cancer, while ERBB3 and ETS2 can be protective markers for good prognosis.

[0103] After integrating the protein-protein interaction network with the CytoHubba algorithm, eight core genes—ERBB3, LGR4, BMP4, CD248, MGP, TGFB3, GRP, and ETS2—were preferentially identified as core regulatory factors in the occurrence of gastric cancer. Among them, TGFB3 showed the highest centrality and significant oncogenic potential, but its research was relatively insufficient in the literature. Therefore, we decided to focus all subsequent computer-aided drug discovery work (including virtual screening, molecular docking, and molecular dynamics simulation) on TGFB3.(Transition)

[0104] 2. Virtual screening of candidate compounds based on the ZINC database

[0105] In this study, we constructed a virtual filtering workflow based on the ZINC database for TGFB3, which sequentially performed LigPrep preprocessing, AutoDockVina preprocessing, and... Maestro's primary screening and Glide's secondary screening evaluated over 1500 commercially available small molecules that met the drug-likeness criteria. Ultimately, the GlideGScores of five compounds were all below -8.3 kcal / mol (Table 1).

[0106] Table 1

[0107]

[0108] Among the top five compounds, ZINC000001530652 (Proflavine hemisulfate) exhibited the best binding energy (–9.323 kcal / mol). Within the binding pocket of TGFB3, this compound stably binds to Tyr178 and Cys180 via π–π stacking interactions and forms a double hydrogen bond network with Glu223, Thr221, and Ser127. Figure 9 A). The second-ranked ZINC000003775644 (Hydroxychloroquine sulfate; –9.215 kcal / mol) forms multi-site stable hydrogen bonds with Glu223, Thr221, and Asn218 at its active site, and establishes tight hydrophobic contacts with hydrophobic residues such as Ile219, Ala217, and Ser125 / 127 through its cyclic backbone. Figure 9 B). The macrocyclic core structure of the third-position ZINC000000005895 (Rizatriptanbenzoate; –8.845 kcal / mol) not only interacts hydrophobically with Phe221, Ile183, and Pro179, but also forms hydrogen bonds with Glu223 and Asp254. Figure 9 C). Fourth position: ZINC000006661227 (L-Histidine; –8.378 kcal / mol; Figure 9 D) and the fifth ZINC000000016154 (Retigabine; –8.327 kcal / mol; Figure 9 E) All of them established 2–3 key hydrogen bonds within the active site, and achieved high complementarity with adjacent polar and hydrophobic residues through their side chain polarity. The above docking results provide ideal candidate molecules for subsequent molecular dynamics simulations and lay a solid theoretical foundation for the design of TGFB3-targeted small molecule drugs.

[0109] 3. Molecular docking to predict the binding ability of TGFB3 to proflavine hemisulfate, hydroxychloroquine sulfate, rizatriptan benzoate, L-histidine and retigabine

[0110] To systematically investigate the interactions between TGFB3 and the five candidate ligands, proflavine hemisulfate, hydroxychloroquine sulfate, rizatriptan benzoate, L-histidine and retigabine, we performed molecular docking using AutoDock Vina. Each ligand was successfully embedded into the TGFB3 binding domain, and the binding affinities were obtained as -9.346, -9.153, -8.489, -8.324 and -8.234 kcal / mol, respectively. In the threshold evaluation of docking energies, a value lower than -4.25 kcal / mol indicates detectable binding activity, a value lower than -5.0 kcal / mol indicates good binding, and a value lower than -7.0 kcal / mol reflects strong binding ability (A-F). These results demonstrate that all five compounds exhibit robust binding interactions with TGFB3. Figure 10

[0111] Among the five compounds identified by virtual screening and molecular docking, proflavine hemisulfate, hydroxychloroquine sulfate, rizatriptan benzoate, L-histidine and retigabine, proflavine hemisulfate showed the lowest binding energy, indicating that its interaction with TGFB3 is the most thermodynamically favorable. Therefore, we selected the TGFB3-proflavine hemisulfate complex to perform all-atom molecular dynamics simulations under explicit solvent conditions. This approach allowed for a detailed evaluation of the conformational stability, key residue interactions, and dynamic behavior of the complex, thereby providing in-depth insights into the mechanism by which proflavine hemisulfate potentially inhibits the activity of TGFB3. (Transition)

[0112] 4. Verification of the binding stability and intermolecular interactions of the TGFB3-proflavine hemisulfate complex

[0113] ​To evaluate the binding stability of TGFB3 with proflavine hemisulfate and its influence on the protein conformation dynamics, we performed 100 ns (100000 ps) molecular dynamics simulation of the complex in Gromacs, and analyzed the RMSD, radius of gyration (Rg), solvent accessible surface area (SASA), principal component free energy landscape, atomic motion covariance matrix and three-dimensional free energy surface in turn.

[0114] The results show that the overall RMSD of the complex rises rapidly after the start of the simulation and converges after about 20 ns, and then fluctuates narrowly in the range of 0.20-0.30 nm in the later stage, indicating that the system has reached equilibrium and maintains good stability Figure 9 A). The total radius of gyration of the complex is maintained between 4.75-5.12 nm, and the axial Rg values also have no sharp fluctuations, reflecting that the molecular conformation is compact and has no obvious loose trend Figure 9 B). During the entire simulation process, the SASA remains in the range of about 900-940 nm 2 , without significant rise or fall, suggesting that the binding of proflavine hemisulfate does not significantly change the surface exposure area of TGFB3 Figure 9 C). The two-dimensional free energy heat map constructed based on principal component analysis (PCA) reveals two main low-energy valleys, with the lowest energy region concentrated at PC1≈0.5, PC2≈4.9, corresponding to ΔG about 2-4 kJ / mol, indicating that the conformation of the complex fluctuates in a limited energy basin Figure 9 D). Covariance matrix: the atomic motion covariance heat map shows that there are obvious positive and negative correlation motions between the residues near the binding pocket, and the highly correlated regions are mainly distributed on the ligand binding surface, further supporting the stabilizing effect of proflavine hemisulfate on the local dynamics of TGFB3 Figure 9 E). The three-dimensional energy map and its bottom projection intuitively show the energy distribution along the PC1 / PC2 direction, and the low-energy potential well region is consistent with the low-energy valley in the two-dimensional free energy landscape, highlighting the binding stable state.

[0115] In summary, the above multi-dimensional simulation analysis consistently shows that the TGFB3 and proflavine hemisulfate complex maintains good conformational stability within 100 ns of simulation time, and the binding pocket residues exhibit ligand-stabilized dynamic characteristics, verifying the tight binding and favorable interaction between the two.

[0116] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for detecting potential protein biomarkers and drug targets for gastric cancer, characterized in that: Includes the following steps: (1) Mendelian randomization analysis: Using genome-wide association study data of gastric cancer from public databases, as well as cis-pQTL data and trans-pQTL data, a two-sample Mendelian randomization (MR) analysis of the proteome was conducted to obtain the causal relationship between plasma proteins and gastric cancer. (2) PPI network construction and key target screening: Protein-protein interaction (PPI) analysis was performed on 29 plasma proteins obtained in step (1) that have a causal relationship with gastric cancer (GC) to screen for biologically significant interactions. Visualize the analysis results and calculate the parameters of each node to select the top-ranked key targets; (3) Validation of key targets: Using single-cell RNA sequencing (scRNA-seq) analysis, high-quality cells were screened, and the cells were clustered and annotated to obtain hub genes. The expression distribution of hub genes in each cell type was analyzed. Kaplan-Meier clinical survival analysis was used to annotate the association between hub gene mRNA levels and the prognosis of gastric cancer patients.

2. The method for detecting novel protein biomarkers and drug targets for gastric cancer according to claim 1, characterized in that: In step (1), the cis-pQTL data and trans-pQTL data are obtained from the deCODE database; the genome-wide association study data of gastric cancer are obtained from the IEGWAS database.

3. The method for detecting novel protein biomarkers and drug targets for gastric cancer according to claim 1, characterized in that: In step (1), in the two-sample Mendelian randomization (MR) analysis, single nucleotide polymorphism (SNP) data are used as instrumental variables (IVs). The single nucleotide polymorphism (SNP) data are mixed plasma protein quantitative trait loci data pQTL, including cis-pQTL data and trans-pQTL data. Single nucleotide polymorphism (SNP) data were screened using the following method: significance threshold P < 5 × 10⁻⁶. -8 Calculate the F-statistic for each SNP, and retain only single nucleotide polymorphism (SNP) data with an F-statistic greater than 10; execute the clumping algorithm using the "TwoSampleMR" R package, with the parameter set to r. 2 =0.001, distance 10,000kb; when single nucleotide polymorphism (SNP) data in the exposure group are missing in the outcome dataset, select the LD(r) with strong similarity to the original SNP. 2 Proxy variants with a value >0.8) are replaced.

4. The method for detecting novel protein biomarkers and drug targets for gastric cancer according to claim 1, characterized in that: In step (1), the two-sample Mendelian randomization (MR) analysis includes: (a) Statistical analysis: performed using R software (version 4.3.2); (b) MR analysis: performed using the "TwoSampleMR" package (Hemanietal., 2018); (c) Inverse variance weighted (IVW) analysis: As the main method for assessing the causal relationship between plasma proteins and gastric cancer, the effect is estimated by weighted regression of SNP-outcome correlation and SNP-exposure correlation, with the intercept assumed to be zero (Burgessetal., 2013). (d) Wald ratio method analysis: When a protein has only a single SNP that meets the requirements, the Wald ratio method is used. (e) False discovery rate (FDR) correction: In order to reduce false positives, the P value was corrected for false discovery rate (FDR), and plasma proteins with a corrected P value of less than 0.05 were considered to be significantly positive; (f) Sensitivity Analysis: To ensure the validity and robustness of the results, sensitivity analysis was employed, including heterogeneity analysis and pleiotropy analysis. Pleiotropy analysis primarily assessed whether horizontal pleiotropy existed among multiple IVs, and the existence of directional pleiotropy was tested using the intercept of MR–Egger regression—when the intercept significantly deviated from zero... (Bowdenetal., 2015) indicates the presence of directed pleiotropy (Bowdenetal., 2019); in heterogeneity analysis, Cochran's Q value is used to measure the degree of difference between IVs, and a high Q value indicates significant variability (Verbancketal., 2018a).

5. The method for detecting novel protein biomarkers and drug targets for gastric cancer according to claim 1, characterized in that: In step (2), specifically: protein-protein interaction (PPI) analysis was performed on plasma proteins causally related to gastric cancer (GC) using the STRING database (https: / / cn.string-db.org / ), with a confidence score threshold of ≥0.4 to screen for biologically significant interactions; the analysis results were visualized, and the parameters of each node were calculated using... Cytoscape software (version 3.10.3, https: / / cytoscape.org / ), with the help of the CytoHubba plugin in Cytoscape, combined with indicators such as Edge Percolated Component (EPC) and Degree (number of node connections), selected the top 10 key targets; and extracted the core targets by cross-filtering the key targets through the construction of Venn graph.

6. The method for detecting novel protein biomarkers and drug targets for gastric cancer according to claim 1, characterized in that: In step (3), specifically: single-cell RNA sequencing (scRNA-seq) analysis is performed using the SeuratR package (Haoetal., 2024) to screen high-quality cells; cells are clustered; and cell types are annotated. The expression distribution of related proteins in different cell types was analyzed.

7. The method for detecting novel protein biomarkers and drug targets for gastric cancer according to claim 6, characterized in that: The high-quality cells must meet the following conditions: mitochondrial gene content is less than 20%, the number of expressed genes exceeds 200, the total number of detected genes is between 200 and 6000, and they must appear in at least three cells; The method for clustering cells is as follows: the Harmony algorithm is used for data integration to correct batch effects and improve clustering accuracy; the data is processed by log-normalization, and the FindVariableFeatures function is used to screen out the top 2000 highly variable genes; then principal component analysis (PCA) is performed to reduce dimensionality, and soft k-means clustering is implemented using the Harmony package; the cells are clustered using the FindClusters function (resolution set to 0.4).

8. The cell type annotation method is based on classical marker genes, differential expression patterns, and known cell characteristics. According to claim 1, a method for detecting novel protein biomarkers and drug targets in gastric cancer is characterized by: The method also includes step (4), which uses Gene Ontology (GO) functional annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis to obtain the potential roles of key genes in biological processes, molecular functions, and / or cellular components, and / or the signaling pathways involved by key genes; in step (4), the Gene Ontology (GO) functional annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis specifically involves: using the online DAVID database (https: / / davidbioinformatics.nih.gov / ) to perform Gene Ontology (GO) functional annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis on gastric cancer-related plasma proteins to obtain the potential roles of key genes in biological processes, molecular functions, and cellular components, as well as the signaling pathways they participate in; and using an online data analysis and visualization platform. (https: / / www.bioinformatics.com.cn), making the results more intuitive.

9. Potential protein biomarkers and drug targets for gastric cancer, characterized by: This includes BMP4, GRP, LGR4, MGP, TGFB3, ERBB3, or ETS2.

10. The application of the method of claim 1 in drug screening for the treatment of gastric cancer, characterized in that: The pivot genes screened using the method of claim 1 are used to screen compounds in the ZINC database using drug virtual screening.

Images