A predictive model for gastric cancer regression after neoadjuvant chemoradiotherapy based on artificial intelligence algorithms

Background: Gastric cancer (GC) is a globally prevalent and lethal malignancy, often diagnosed at locally advanced stages. Although neoadjuvant chemoradiotherapy (CRT) is a key multimodal treatment, its variable efficacy underscores the urgent need for predictive biomarkers and mechanistic exploration.

Methods: A total of 20 GC patients who underwent preoperative CRT followed by surgery between 2018 and 2020 were retrospectively enrolled in this study. Endoscopic biopsy specimens collected preoperatively from these patients were retrieved for transcriptome sequencing. Patients were divided into positive-response and negative-response groups based on cTNM/pTNM staging. LASSO, Support Vector Machine, Random Forest algorithms, and an artificial intelligence neural network model were used for data analysis, core gene screening, and predictive model construction. GSEA and immune cell infiltration analysis were performed to explore mechanisms. The expression patterns of identified genes were validated in vitro using oxaliplatin-resistant and wild-type GC cell lines via qPCR.

Results: We identified 1224 differentially expressed genes (DEGs) between the two groups. Six key genes (RPL17, GCLM, HOXC10, USP18, PKN1, NUP93) were screened via machine learning algorithms. The AI neural network model based on these genes showed perfect predictive efficacy (AUC = 1.000). The negative-response group had higher neutrophil infiltration. PKN1 and RPL17 are positively correlated with M2 cell expression in the positive-response group, while they show a negative correlation with M1 cell expression in the negative- response group. GSEA indicated significant enrichment of DEGs in immune response, metabolism, and ribosome pathways. qPCR experiments revealed that the tissue expression levels of the core genes in the negative response group are consistent with the expression trends observed in gastric cancer drug-resistant cell lines.

Conclusion: The six-gene combination is a promising predictive biomarker for GC neoadjuvant CRT efficacy, providing a basis for personalized treatment. Future prospective multi-center studies and multimodal evidence are needed to validate the model and elucidate the underlying mechanisms.