Machine Learning-Driven Prediction Model for Cuproptosis-Related Genes in Spinal Cord Injury: Construction and Experimental Validation

Yimin Zhou ¹ Xin Li ² Zixiu Wang ³ Liqi Ng ⁴ He Rong ⁵ Chaozong Liu ⁴ Gang Liu ¹ Xiao Fan ⁶ Xiaohong Mu ¹ Yu Zhou ^2,7*

• ¹ Department of Orthopedics, Dongzhimen Hospital, Beijing University of Chinese Medicine, Beijing, Beijing Municipality, China
• ² Postdoctoral research workstation, Chongqing Orthopedic Hospital of Traditional Chinese Medicine, Chongqing, China
• ³ College of Pharmacy, Gannan Medical University, Ganzhou, Jiangxi Province, China
• ⁴ Institute of Orthopaedic and Musculoskeletal Science, University College London, London, England, United Kingdom
• ⁵ College of Integrated Chinese and Western Medicine, Changchun University of Chinese Medicine, Changchun, Jilin Province, China
• ⁶ Department of Orthopedics, Qingdao Municipal Hospital, Qingdao, Shandong Province, China
• ⁷ Department of Orthopedics, First Affiliated Hospital of Chongqing Medical University, Chongqing, Chongqing Municipality, China

Spinal cord injury (SCI) represents a severe disorder of the central nervous system that causes sensory and motor dysfunction. As an essential trace element in living organisms, copper plays a vital role in life activities. Cuproptosis is a novel metal ion-induced cell death distinct from ferroptosis and closely related to mitochondrial metabolism. In this study, we conducted a comprehensive bioinformatics analysis of the SCI (GSE151371) dataset to explore the variations in expression and correlations of Cuproptosis-related genes (CRGs) between SCI and non-SCI and used the ssGSEA algorithm for immune infiltration analysis. In addition, unsupervised clustering was performed based on CRGs, and WGCNA, GO, and KEGG enrichment analyses were performed based on the clustering results. Then, three machine learning models were constructed, the best candidate genes were selected, and finally, the nomogram model and qRT-PCR experiments were used for verification. The results showed that there were 7 CRGs differentially expressed, and there were significant differences in immune cells such as Activated.dendritic.cell and Macrophage. Cluster C1 and Cluster C2 were obtained by unsupervised clustering, and three key module genes between clusters were obtained by WGCNA involving related pathways such as Ribosome and HIF-1 signaling pathway. Four candidate genes (SLC31A1, DBT, DLST, LIAS) were obtained from RF, Lasso, and SVM models, of which three were down-regulated, and one was up-regulated, and the SLC31A1 performance was the best (AUC = 0.958). The Nomogram model, ROC curve, and experimental results proved that the four candidate genes were highly reliable. Therefore, this study explored the association between SCI and cuproptosis based on the GEO dataset and obtained highly reliable prediction models and genes through a variety of machine learning methods, which is expected to provide new clinical evidence and support for potential therapeutic targets in SCI.