Abstract:

Objective　To screen the ferroptosis-related genes (FRGs) in sepsis by public databases and machine learning algorithms and explore their roles in order to provide theoretical basis for diagnosis and treatment of sepsis. Methods　Samples of sepsis patients were obtained from the GEO database, including GSE185263 as the training dataset and GSE154918 as the external validation dataset. Differentially expressed genes (DEGs) associated with ferroptosis were firstly obtained and followed by biofunctional analysis, such as gene ontology (GO) enrichment analysis, Kyotoencyclopedia of genes and genomes (KEGG) pathway analysis, and disease ontology (DO) enrichment analysis. The LASSO regression and SVM-RFE machine-learning algorithms were then used to screen the FRGs. In addition, the construction of a nomogram was based on the expression content of the obtained genes. Finally, the differences in immune cell infiltration between the normal control group and sepsis group were analyzed. Results　A total of 87 ferroptosis-related DEGs were obtained. GO enrichment analysis showed that the biological processes of these DEGs were mainly enriched in oxidative stress, regulatory pathways of apoptotic signaling, cellular senescence, and regulation of neuronal death; cellular components were enriched in autophagosomes, nuclear envelope, and secondary lysosomes; molecular functions were enriched in kinase activity of mitogen-activated protein kinase (MAP kinase) and iron binding. The KEGG pathways were mainly enriched in the FoxO signaling pathway, hypoxia-inducible factor (HIF)-1 signaling pathway, PI3K-Akt signaling pathway, and ferroptosis pathway. DO analysis showed that sepsis was closely related to neurological tumors, neuroblastoma, bone cancer, and other diseases. Machine learning showed that PRDX1, IDH1, DUSP1, YWHAE, and SOCS1 presented as the hub FRGs, and these genes and their nomogram model all had high diagnostic values for sepsis. The immune infiltration results showed that the activities of naive B cells, CD8+ T cells, CD4+ memory resting T cells, follicular helper T cells, resting natural killer (NK) cells, resting dendritic cells, active dendritic cells were suppressed in the sepsis group, and plasma cells, naive T cells, active memory CD4+ T cells, δ T cells, macrophages M0, and neutrophils were active (all P<0.05). Conclusion　PRDX1, IDH1, DUSP1, YWHAE, and SOCS1 are all potentially ferroptosis-related biomarkers for sepsis diagnosis.

Key words:

Sepsis, Ferroptosis, Bioinformatics, Machine learning

摘要：

目的　利用公共数据库中相关数据联合机器学习算法筛选脓毒血症中铁死亡相关基因，并分析其作用的理论依据，为脓毒血症的诊疗提供理论依据。方法　从GEO数据库获取脓毒血症患者的样本，其中GSE185263作为训练数据集分析，GSE154918作为外部验证数据集。先获取铁死亡相关的差异表达基因（differentially expressed genes，DEGs），并进行基因本论（gene ontology，GO）富集分析、京都基因与基因组百科全书（Kyotoencyclopedia of genes and genomes，KEGG）通路分析和疾病本体（disease ontology，DO）富集分析。再运用LASSO回归、SVM-RFE机器学习算法筛选铁死亡相关基因，并基于获得基因的表达含量，构建列线图用以预测脓毒血症。最后，分析正常对照组和脓毒血症组之间的免疫细胞浸润差异。结果　共获取87个铁死亡相关的DEGs，GO富集显示生物学过程主要富集在氧化应激、凋亡信号的调控路径、细胞衰老、神经元死亡的调节等方面；细胞组分富集在自噬体、核包膜和二级溶酶体等方面；分子功能富集在促分裂原活化蛋白激酶（MAP激酶）的激酶活性、铁离子结合等方面。KEGG通路主要富集在FoxO信号传导途径、缺氧诱导因子（HIF）-1信号传导途径、PI3K-Akt信号传导途径和铁死亡等通路。DO分析结果显示脓毒血症同神经系统肿瘤、神经母细胞瘤、骨癌等疾病关系密切。机器学习筛选出PRDX1、IDH1、DUSP1、YWHAE和SOCS1为铁死亡相关核心基因，上述基因和列线图模型对于脓毒血症均具有较高的诊断价值。免疫浸润结果显示，脓毒血症组中幼稚B细胞、CD8+ T细胞、CD4+记忆静止T细胞、滤泡辅助性T细胞、静止自然杀伤（NK）细胞、静止树突状细胞、活动期树突状细胞活动受到抑制，浆细胞、幼稚T细胞、活动期记忆CD4+ T细胞、δ T细胞、巨噬细胞M0、中性粒细胞活跃（均P<0.05）。结论　PRDX1、IDH1、DUSP1、YWHAE和SOCS1均为潜在的脓毒血症诊断相关的铁死亡生物标志物。

关键词:

脓毒血症, 铁死亡, 生物信息学, 机器学习