Gastric cancer (GC) remains a major global malignancy characterized by dysregulation of multiple signaling pathways and frequent alterations in tumor suppressor genes. Tumor suppressor proteins regulate cell-cycle control, DNA repair, and apoptosis; pathogenic genetic variations in these genes can abrogate their protective functions and promote tumorigenesis. Although many individual gene studies exist, a comprehensive systems-level approach that integrates expression dysregulation with mutation frequency to prioritize clinically relevant tumor suppressor variations in GC was lacking. This study addresses that gap by computationally screening genes and variants that both change expression in GC and show high mutation propensity, with a focus on the tumor suppressors TP53, CDH1, and APC.
The authors implemented a deep learning approach based on a graph neural network to detect genes that simultaneously exhibit differential expression and mutation tendency in GC. Model performance metrics were reported: mean squared error (MSE) ranged from 0.00482 ± 0.00023 to 0.07108 ± 0.00437; R² from 0.85098 ± 0.01903 to 0.85899 ± 0.01987; and AUC-ROC from 0.93095 ± 0.01758 to 0.93309 ± 0.00725. Using this approach, the pipeline prioritized 1,886 genes that satisfy both criteria of dysregulated expression and mutation propensity in GC.
Graph neural network–based network analysis identified central hub genes among the prioritized list. TP53 emerged as the most prominent hub with a reported mutation frequency of 47.6%. Other notable hubs included ERBB2 (8.8%), CDH1 (8.2%), and APC (6.8%). These hubs were targeted for in-depth computational characterization because of their known roles as tumor suppressors or key regulatory oncogenes in gastric tumorigenesis.
The study performed a multilayer computational interrogation of TP53, CDH1, and APC, encompassing evolutionary conservation profiling, biophysical energetics assessment, and conformational-change detection using unsupervised machine learning. These analyses aimed to predict how specific genetic variants alter protein stability, structure, and function, and to identify variants likely to disrupt tumor-suppressive activity. Details of individual variant-level biophysical metrics and conservation scores are reported within the paper and supporting files.
Variants identified computationally were validated against patient-derived data in cBioPortal. Validation confirmed an association of GC with 36 missense single-nucleotide polymorphisms (SNPs) that critically affect post-translational modification sites, specifically methylation and phosphorylation sites, and with 60 nonsense SNPs. The authors highlight that missense SNPs affecting such modification sites can have complex consequences on protein function beyond simple amino-acid substitution, by altering regulatory modifications that modulate activity, stability, or interactions.
Expression analysis indicated that TP53, CDH1, and APC were significantly upregulated in GC tissues relative to control samples. These expression changes were associated with altered patient survival; the manuscript reports statistically significant survival associations (p < 0.05). The directionality and clinical effect sizes are presented in the article’s figures and supporting tables.
The analysis identified shared regulatory mechanisms that may modulate the three tumor suppressors. Specifically, the transcription factor EZH2 and microRNA miR-129-5p were reported as common regulatory elements affecting TP53, CDH1, and APC. Additionally, mutations in these tumor suppressors were associated with dysregulation of multiple oncogenes, including CCNE1/2 and FGFR2, suggesting that pathogenic variants propagate perturbations across GC-relevant signaling networks.
By integrating deep learning prioritization, network topology, variant-level computational biophysics, and clinical validation, the study constructs a systems-level molecular framework. This framework links pathogenic variants in key tumor suppressors to loss of tumor-suppressive function and downstream oncogene activation, and it highlights candidate biomarkers and potential molecular targets relevant for precision diagnostics or therapeutic decision-making in GC. The authors propose that identification of variants affecting post-translational modification sites and shared regulatory elements may inform more precise stratification of patients.
The study demonstrates a computational pipeline that identifies genes and pathogenic variants with dual evidence of expression dysregulation and mutation propensity in gastric cancer, with TP53 prioritized as the highest-frequency hub. Validated variant sets include missense SNPs affecting methylation and phosphorylation sites and a set of nonsense SNPs linked to GC. Shared regulators (EZH2 and miR-129-5p) and downstream oncogene dysregulation were also reported. All relevant data and supporting information are provided in the paper and its supporting files. The authors declared no specific funding and no competing interests.
Note: This summary and rewrite follow details reported in the source article. Specific per-variant scores, full lists of the 1,886 genes, and detailed biophysical metrics are available in the article’s figures and supporting information files.