机器学习识别肝母细胞瘤核心驱动基因及免疫细胞浸润分析

Machine learning to identify core driver genes of hepatoblastoma and analysis of immune cell infiltration

目的通过机器学习识别肝母细胞瘤(hepatoblastoma,HB)的核心驱动基因,并探讨HB免疫细胞浸润特征及与核心驱动基因关系。方法从基因表达合成数据库(gene expression omnibus,GEO)中下载3套数据集(GSE75271、GSE131329和GSE81928),通过最小绝对收缩和选择操作符(least absolute shrinkage and selection operator,LASSO)和支持向量机(support vector machine,SVM)-回归特征消除(recursive feature elimination,RFE)算法识别核心驱动基因。在验证集中,对核心驱动基因进行表达和诊断验证。使用CIBERSORT算法评估HB组织的免疫细胞浸润特征,并评估免疫细胞浸润与核心驱动基因之间相关性。采用实时聚合酶链反应(real-time polymerase chain reaction,RTPCR)和蛋白免疫印迹对3对HB组织和肝正常组织样本进行核心驱动基因验证。结果LASSO回归及SVM-RFE算法识别5个核心驱动基因:RDH16、EPCAM、CYP1A2、MGLL和SLC27A5。测试集中,HB组织样本RDH16、CYP1A2、MGLL和SLC27A5表达量低于肝正常组织样本(P<0.05),EPCAM高于肝正常组织样本(P<0.05),RDH16、EPCAM、CYP1A2、MGLL和SLC27A5在诊断HB的受试者工作特征曲线下面积分别为0.992,0.990,0.980,0.993,0.987。验证集中的结果与测试集结果一致。PCR结果显示,HB组织样本RDH16、CYP1A2、MGLL和SLC27A5表达量低于肝正常组织样本(P<0.05),EPCAM高于肝正常组织样本(P<0.05);蛋白免疫印迹结果与PCR结果一致。免疫细胞浸润分析显示5个核心驱动基因表达与活化肥大细胞、中性白细胞、T细胞CD8等免疫细胞绝对含量相关。结论RDH16、EPCAM、CYP1A2、MGLL和SLC27A5是HB核心驱动基因,免疫细胞浸润之间交互作用表明这些核心驱动基因可能成为未来治疗HB的一个潜在靶点。

Objective To identify the core driver genes of hepatoblastoma(HB)through machine learning and explore the characteristics of HB immune cell infiltration and examine the relationship with core driver genes.Methods Three sets of data(GSE75271,GSE131329&GSE81928)were downloaded from the Gene Expression Synthesis Database(GEO).And the least absolute shrinkage and selection operator(LASSO)and support vector machine(SVM)-regression feature elimination(RFE)algorithm were employed for identifying the core driver genes.The expression and diagnosis of core driver genes were verified in the verification set.The CIBERSORT algorithm was utilized for evaluating the characteristics of immune cell infiltration in HB tissue and elucidating the correlation between immune cell infiltration and core driver genes.Polymerase chain reaction(PCR)and Western blot were employed for verifying the core driver genes of 3 pairs of HB tissue and normal liver tissue samples.Results LASSO regression and SVM-RFE algorithm revealed five core driving genes of RDH16,EPCAM,CYP1A2,MGLL and SLC27A5.In the test set,the expression levels of HB tissue samples RDH16,CYP1A2,MGLL and SLC27A5 were lower than normal liver tissue samples(P<0.05)and EPCAM was higher than normal liver tissue samples(P<0.05).RDH16,EPCAM,CYP1A2,MGLL and SLC27A5 were in area under curve of receiver operating characteristic(ROC)for diagnosing HB were 0.992,0.990,0.980,0.993 and 0.987 respectively.The results in the verification set were consistent with the results in the test set.Polymerase chain reaction(PCR)results showed that the expression levels of RDH16,CYP1A2,MGLL and SLC27A5 were lower in HB tissue samples than those in normal liver tissue samples(P<0.05)and EPCAM was higher than normal liver tissue samples(P<0.05);Western blot results were consistent with PCR results.Immune cell infiltration analysis showed that the expressions of 5 core driving genes were correlated with the absolute content of immune cells,such as activated mast cells,neutrophils,CD8+T cells.Conclusions RDH16,EPCAM,CYP1A2,MGLL and SLC27A5 are the core driving genes of HB.The interaction between immune cell infiltration implies that these core driver genes may be a potential target for the future treatment of HB.