Table 2_Identifying metabolism-related genes in liver cancer through weighted gene co-expression network analysis and machine learning.xls

ObjectiveAs a leading cause of cancer-related mortality, liver cancer was associated with metabolic dysregulation. We aimed to identify metabolism-related prognostic biomarkers and therapeutic targets.
MethodsTranscriptomic data from TCGA were analyzed using EdgeR to identify differentially expressed genes (DEGs). WGCNA was applied to unveil the metabolism-related genes in liver cancer. Machine learning algorithms (RF, SVM, LASSO) refined marker genes. GSEA and ssGSEA were conducted to identify pathway associations and immune interactions of marker genes. DGIdb database predicted candidate therapeutics targeting these biomarkers. The independent queue (GSE54236) was verified as an external dataset. RT-PCR validated gene expression in clinical samples.
ResultsA total of 234 metabolism-related genes were identified in liver cancer. Through undergoing machine learning by RF, SVM, and LASSO algorithms, seven marker genes (ACADS, ALDH8A1, COX4I2, CYP2C8, DBH, NDST3, and PLA2G6) were obtained. Except for PLA2G6, the other genes were correlated with the survival of patients with liver cancer and immune cells infiltration. Additionally, ACADS, ALDH8A1, CYP2C8, DBH, and NDST3 were downregulated, and COX4I2 was upregulated in dataset of GSE54236, which were consist with those in TCGA database. However, RT-PCR validation in 10 paired clinical samples confirmed significant downregulation of ACADS, ALDH8A1, COX4I2, CYP2C8, DBH, and NDST3 in tumor tissues (all P < 0.05). Immune infiltration analysis revealed these genes might influence immune cell infiltration in the tumor microenvironment. And the candidate drugs were unveiled, including PAZOPANIB, SUMATRIPTAN, ETOPOSIDE, etc.
ConclusionThe metabolism-related biomarkers ACADS, ALDH8A1, COX4I2, CYP2C8, DBH, and NDST3 demonstrated significant potential for predicting liver cancer prognosis and may serve as candidate therapeutic targets.

目的 肝癌作为癌症相关死亡的首要诱因，与代谢失调密切相关。本研究旨在筛选代谢相关的预后生物标志物及治疗靶点。
方法 本研究通过EdgeR软件分析癌症基因组图谱（TCGA, The Cancer Genome Atlas）中的转录组数据，以鉴定差异表达基因（DEGs, differentially expressed genes）。采用加权基因共表达网络分析（WGCNA, Weighted Gene Co-expression Network Analysis）挖掘肝癌组织中与代谢相关的基因。借助机器学习算法（随机森林RF、支持向量机SVM、最小绝对收缩和选择算子LASSO）进一步筛选标记基因。通过基因集富集分析（GSEA, Gene Set Enrichment Analysis）及单样本基因集富集分析（ssGSEA, single-sample Gene Set Enrichment Analysis）探究标记基因的通路关联与免疫互作特征。利用DGIdb数据库预测靶向这些生物标志物的候选治疗药物。以独立队列GSE54236作为外部验证数据集开展验证，并通过实时荧光定量PCR（RT-PCR, reverse transcription polymerase chain reaction）检测临床样本中的基因表达水平。
结果 本研究共在肝癌组织中鉴定得到234个代谢相关基因。经RF、SVM及LASSO三种机器学习算法筛选后，最终获得7个标记基因：ACADS、ALDH8A1、COX4I2、CYP2C8、DBH、NDST3及PLA2G6。除PLA2G6外，其余基因均与肝癌患者的生存预后及免疫细胞浸润水平显著相关。在GSE54236数据集的验证中，ACADS、ALDH8A1、CYP2C8、DBH及NDST3呈下调表达，COX4I2呈上调表达，该结果与TCGA数据库中的分析结果一致。但针对10对临床样本的RT-PCR验证结果显示，ACADS、ALDH8A1、COX4I2、CYP2C8、DBH及NDST3在肿瘤组织中均显著下调（所有P < 0.05）。免疫浸润分析表明，上述标记基因可通过影响肿瘤微环境中的免疫细胞浸润发挥调控作用。本研究同时筛选得到潜在候选药物，包括帕唑帕尼（PAZOPANIB）、舒马曲坦（SUMATRIPTAN）、依托泊苷（ETOPOSIDE）等。
结论 本研究筛选得到的代谢相关生物标志物ACADS、ALDH8A1、COX4I2、CYP2C8、DBH及NDST3，在肝癌预后预测中具有显著潜力，可作为潜在的治疗靶点。