背景:同源重组在胃癌的发生和耐药中起着至关重要的作用。本研究旨在筛选同源重组途径中新的胃癌诊断生物标志物,然后利用影像组学特征构建生物标志物表达预测模型,指导化疗方案的选择。
方法:从癌症基因组图谱数据库下载胃癌转录组数据。使用机器学习方法来筛选胃癌的诊断生物标志物并通过实验验证它们。胃癌患者的计算机断层扫描图像数据和相应的临床数据从癌症影像档案和我们的成像中心下载,然后对计算机断层扫描图像进行特征提取,构建生物标志物表达预测模型,分析生物标志物影像组学评分与临床病理特征的相关性。
结果:我们通过机器学习筛选了同源重组途径中的RAD51D和XRCC2作为胃癌诊断的生物标志物,RAD51D和XRCC2的表达与病理T分期呈显著正相关,N级,TNM阶段。同源重组通路阻断抑制胃癌细胞增殖,促进细胞凋亡,降低胃癌细胞对化疗药物的敏感性。我们的预测RAD51D和XRCC2表达模型是使用影像组学特征构建的,所有模型都有很高的准确性。在外部验证队列中,预测模型仍然具有不错的准确性。此外,RAD51D和XRCC2的影像组学评分也与病理T呈显著正相关,N,和TNM阶段。
结论:我们筛选的胃癌诊断生物标志物RAD51D和XRCC2可以,在某种程度上,通过放射学特征反映基因的表达状态,对胃癌患者化疗方案的选择具有一定的指导意义。
BACKGROUND: Homologous recombination plays a vital role in the occurrence and drug resistance of gastric cancer. This study aimed to screen new gastric cancer diagnostic biomarkers in the homologous recombination pathway and then used radiomic features to construct a prediction model of biomarker expression to guide the selection of chemotherapy regimens.
METHODS: Gastric cancer transcriptome data were downloaded from The Cancer Genome Atlas database. Machine learning methods were used to screen for diagnostic biomarkers of gastric cancer and validate them experimentally. Computed Tomography image data of gastric cancer patients and corresponding clinical data were downloaded from The Cancer Imaging Archive and our imaging centre, and then the Computed Tomography images were subjected to feature extraction, and biomarker expression prediction models were constructed to analyze the correlation between the biomarker radiomics scores and clinicopathological features.
RESULTS: We screened
RAD51D and XRCC2 in the homologous recombination pathway as biomarkers for gastric cancer diagnosis by machine learning, and the expression of
RAD51D and XRCC2 was significantly positively correlated with pathological T stage, N stage, and TNM stage. Homologous recombination pathway blockade inhibits gastric cancer cell proliferation, promotes apoptosis, and reduces the sensitivity of gastric cancer cells to chemotherapeutic drugs. Our predictive
RAD51D and XRCC2 expression models were constructed using radiomics features, and all the models had high accuracy. In the external validation cohort, the predictive models still had decent accuracy. Moreover, the radiomics scores of
RAD51D and XRCC2 were also significantly positively correlated with the pathologic T, N, and TNM stages.
CONCLUSIONS: The gastric cancer diagnostic biomarkers
RAD51D and XRCC2 that we screened can, to a certain extent, reflect the expression status of genes through radiomic characteristics, which is of certain significance in guiding the selection of chemotherapy regimens for gastric cancer patients.