【作者】 袁春艳；

【导师】 王开发；

【作者基本信息】 西南大学 ， 应用统计（专业学位）， 2022， 硕士

【摘要】 癌症在世界范围内已成为引起人们死亡的最主要因素之一,其死亡率高且治愈率较低。它的产生是多种因素相互作用的结果。而胃癌作为最常见的癌症之一,其早期症状不明显且一般治愈率也较低,目前对它的治疗方案大多是临床手术治疗。随着基因芯片技术和高通量测序技术的产生和发展,已经得到海量基因表达数据,能够帮助人们从中挖掘出大量有用的信息,这给研究者们提供了一种挖掘新型的分子靶向治疗方案的可能,极大地促进了对癌症的预防和诊断治疗。现今综合统计方法和机器学习算法对胃癌的基因表达谱数据进行挖掘分析,期望能筛选出对胃癌的预后和诊断有影响的特征基因。从在TCGA(The Cancer Genome Atlas)数据库提取到了32例正常样本和375例胃癌样本的基因表达数据、以及443例胃癌患者的临床数据,经ID号比对和缺失数据的删除,其中既有基因表达数据也有临床数据的患者共368例。经过数据预处理后,利用R语言软件自带的差异分析功能包edge R、DESeq2、limma对基因进行挑选,设置挑选标准(|log FC|>1和p<0.05),结果发现在同等条件下edgeR、DESeq2挑选的差异基因log FC值和P值差别不大,limma包挑选的差异基因更精确。最终选取了三个包挑选出的1726个交集基因作为差异基因进行下一步分析。对挑选出来的1726个差异基因,分别建立XGboost和决策树分类模型进行正常样本和癌症样本的分类预测,并对比两个模型的在训练集和测试集ROC曲线。结果发现决策树分类模型的预测效果优于XGboost模型,最后选取决策树分类模型挑选出的11个基因作为最优子集,并利用GO和KEGG方法对这些基因进行富集研究,以进一步认识其在人体中的功能。将11个特征基因的表达量数据和临床数据进行结合,首先对临床数据中的癌症患者的性别、年龄等相关临床数据与基因表达水平进行t检验,找出与临床参数相关的基因。其次对基因的表达水平与生存状态(futime、fustat)之间的关系进行单因素Cox生存分析和多因素Cox分析,结果显示在单因素Cox分析中得到4个与生存预后相关的基因:NFE2L3、MIR4435-2HG、CTHRC1、HBB;在多因素Cox分析中得到3个与胃癌患者生存状态有影响的预后基因:NFE2L3、CTHRC1、HBB。综上,文章对胃癌患者的基因表达谱数据进行差异分析选取了差异基因、分类预测选取了特征基因、生存分析选取了预后基因。结果显示根据特征基因建立的分类预测模型能很好的对癌症-正常进行预测,生存分析选取的预后基因也能显著影响癌症患者的生存状态,并在不同的通路有不同的功能,最后经过文献查阅也验证筛选出的预后基因对胃癌患者确实有相关影响,相关结论对胃癌的诊断和预后研究有一定的参考价值。更多还原

【Abstract】 Cancer has become one of the leading causes of death worldwide,with high mortality and low cure rates.Its production is the result of the interaction of many factors.As one of the most common cancers,gastric cancer has no obvious early symptoms and generally has a low cure rate.At present,most of the treatment options for it are clinical surgery.With the generation and development of gene chip technology and high-throughput sequencing technology,a large amount of gene expression data has been obtained,which can help people to mine a lot of useful information,which provides researchers with a new molecular targeted therapy for mining.The possibility of the program has greatly promoted the prevention and diagnosis and treatment of cancer.Nowadays,comprehensive statistical methods and machine learning algorithms are used to mine and analyze the gene expression profile data of gastric cancer,hoping to screen out the characteristic genes that have an impact on the prognosis and diagnosis of gastric cancer.The gene expression data of 32 normal samples and 375 gastric cancer samples,and the clinical data of 443 gastric cancer patients were extracted from the TCGA(The Cancer Genome Atlas)database.After ID number comparison and deletion of missing data,existing genes A total of 368 patients had clinical data for expression data.After data preprocessing,genes were selected using the differential analysis function packages edge R,DESeq2,and limma that come with R language software,and the selection criteria(|log FC| >1 and p<0.05)were set.It was found that under the same conditions,edge R,The log FC value and P value of the differential genes selected by DESeq2 are not much different,and the differential genes selected by the limma package are more accurate.Finally,1726 intersecting genes selected by three packages were selected as differential genes for further analysis.For the selected 1726 differential genes,XGboost and decision tree classification models were established to predict the classification of normal samples and cancer samples,and the ROC curves of the two models in the training set and test set were compared.The results showed that the prediction effect of the decision tree classification model was better than that of the XGboost model.Finally,11 genes selected by the decision tree classification model were selected as the optimal subset,and GO and KEGG methods were used to enrich these genes to further understand their genes.function in the human body.The expression data of 11 characteristic genes and clinical data were combined.First,t-test was performed on the clinical data and gene expression levels of cancer patients in the clinical data,such as gender and age,to find out the genes related to clinical parameters.Secondly,univariate Cox survival analysis and multivariate Cox analysis were performed on the relationship between gene expression level and survival status(futime,fustat).MIR4435-2HG,CTHRC1,HBB;Three prognostic genes affecting the survival status of gastric cancer patients were obtained in multivariate Cox analysis: NFE2L3,CTHRC1,HBB.To sum up,this thesis selects differential genes for differential analysis of the gene expression profile data of gastric cancer patients,selects characteristic genes for classification prediction,and selects prognostic genes for survival analysis.The results show that the classification prediction model established based on characteristic genes can predict cancer-normal well,and the prognosis genes selected by survival analysis can also significantly affect the survival status of cancer patients,and have different functions in different pathways.The review also verified that the selected prognostic genes do have relevant effects on gastric cancer patients,and the relevant conclusions have certain reference value for the diagnosis and prognosis of gastric cancer.更多还原