Machine Learning Approach Identifies a Novel Ferroptosis‐Related Gene Model in Aberrant DNA Methylation for Predicting the Progressive Risk of Primary Myelofibrosis

Introduction：Primary myelofibrosis (PMF) is clonal hematopoietic disorder including aberrant DNA methylation, carrying a high risk of progression to acute myeloid leukemia (AML). Ferroptosis is a novel form of iron-dependent cell death driven by lipid peroxidation, and recent studies have confirmed that ferroptosis plays an important role in the development of AML. However, the ferroptosis-related genes (FRGs) in PMF progression remain to be discovered.

Methods：PMF‐related microarray data (GSE42042, GSE152519) were downloaded from the NCBI Gene Expression Omnibus (GEO) database. The LIMMA package was used to identify differentially expressed genes (DEGs) in PMF patients with and without progression, followed by the intersection of DEGs and FRGs (FRG-DEGs). Functional enrichment analysis, protein-protein interaction (PPI) network construction, and machine learning algorithms (least absolute shrinkage and selection operator (LASSO) and random forest) were applied to identify hub genes. A prediction model was established by generated a nomogram and receiver operating characteristic curve (ROC) for the progress risk of PMF patients with aberrant DNA methylation. The predictive ability was identified by an external data set and our clinical PMF samples. Immune cell infiltrations were analyzed, and functional analysis was applied to explore the underlying mechanisms. The mRNA levels of model genes were verified in PMF clinical samples by quantitative real‐time polymerase chain reaction (qRT‐PCR).

Results：A total of 1603 DEGs were identified between PMF and PMF_sAML groups in the dataset of GSE42042, of which 1758 were downregulated, and 25 were upregulated. 41 FRG-DEGs were identified with intersection of DEGs and FRGs (Figure 1A). 22 FRG-DEGs were selected using venn plots for further machine learning analysis (Figure 1B). Finally, five hub genes (NRAS, KRAS, PIK3CA, CHMP6, ELOVL5) were identified and used to establish a nomogram that yielded a high predictive performance. The area under the curve (AUC) of predictive model in training and validation cohorts were 1.000 and 0.872, respectively. In addition, 6 patients with PMF progression and 13 patients with stable PMF were enrolled, and their samples of bone marrow were detected with RNA sequencing in our clinaical center of Zhejiang University School of Medicine. The AUC of predictive model in our clinical data also yielded a high predictive performance with 1.000 (Figure 1C). Compared to PMF stable group, the mRNA expression levels of NRAS, KRAS, PIK3CA, CHMP6, and ELOVL5 were all upregulated in PMF progressive group (Figure 1D). KEGG analysis revealed that these five hub genes were significantly involved in VEGF signaling pathway, longevity regulating pathway, Fc epsilon RI signaling pathway, PD−L1 expression and PD−1 checkpoint pathway, and B cell receptor signaling pathway. Gene ontology (GO) functional enrichment analysis showed that five hub genes were enriched in epidermal growth factor receptor (EGFR) signaling, positive regulation of gene expression, fatty acid elongase activity, 1-phosphatidvlinosito|-4-phosphate 3-kinase activity(Figure 1F). Moreover, we found a positive correlation between the five hub DEGs and immune cell infiltrations, including Moncytes, M0 nad M2 Macrophages cells, T cell gamma delta(Figure 1G).

Conclusion：Our study identified five hub genes (NRAS, KRAS, PIK3CA, CHMP6, ELOVL5), and established a novel ferroptosis‐related gene model using machine learning algorithm to well predict the progressive risk of PMF. FRGs may be implicated in PMF pathogenesis through immune‐related pathways, and aid in identifying potential therapeutic targets for PMF patients.

This research was funded by the Key R&D Program of Zhejiang (No. 2022C03137) and the Zhejiang Medical Association Clinical Medical Research special fund project (No. 2022ZYC-D09). *Correspondence to: Prof Jian Huang, E-mail: househuang@zju.edu.cn.

No relevant conflicts of interest to declare.