Revealing the novel ferroptosis-related therapeutic targets for diabetic foot ulcer based on the machine learning

<b>Objectives:</b> DFU is a serious chronic disease with high disability and fatality rates, yet there is no completely effective therapy. While ferroptosis is integrated to inflammation and infection, its involvement in DFU is still unclear. The study aimed to identify ferroptosis-related genes in DFU, providing potential therapeutic targets. <b>Methods:</b> In the GEO database, two DFU microarray datasets (GSE147890 and GSE80178) were collected. WGCNA was conducted to identify the modular genes most involved in DFU. Subsequently, enrichment analysis and PPI analysis were performed. To yield the DFU-associated ferroposis genes, the ferroposis genes were retrieved from the FerrDb database and overlapped with the modular genes. Eventually, an optimal DFU prediction model was created by combining multiple machine learning algorithms (LASSO, SVM-RFE, Boruta, and XGBoost) to detect ferroposis genes most closely associated with DFU. The accuracy of the model was verified by utilizing external datasets (GSE7014) based on ROC curves. <b>Results:</b> WGCNA yielded seven modules in all, and 1223 DFU-related modular genes were identified. GO analysis revealed that inflammatory response, decidualization, and protein binding were the most highly enriched terms. These module genes were also enriched in the <i>ErbB</i> signaling, <i>IL-17</i> signaling, <i>MAPK</i> signaling, growth hormone synthesis, secretion and action, and tight junction KEGG pathways. Twenty-five DFU-associated ferroposis genes were obtained by cross-linking with modular genes, which could distinguish DFU patients from controls. Ultimately, the prediction model based on machine learning algorithms was well established, with high AUC values (0.79 of LASSO, 0.80 of SVM, 0.75 of Boruta, 0.70 of XGBoost). <i>MAFG</i> and <i>MAPK3</i> were identified by the prediction model as the most highly associated ferroposis-genes in DFU. Furthermore, the external dataset (GSE29221) validation revealed that <i>MAPK3</i> (AUC = 0.81) had superior AUC values than <i>MAFG</i> (AUC = 0.62). <b>Conclusion:</b> As the most related ferroptosis-genes with DFU, <i>MAFG</i> and <i>MAPK3</i> may be employed as potential therapeutic targets for DFU patients. Moreover, <i>MAPK3</i>, with higher accuracy, could be the more potential ferroptosis-related biomarker for further experimental validation.