大腸がんにおける主要なクロトーシス関連標的の特定：単一細胞RNAシーケンシングの機械学習およびAUcell解析によるトランスクリプトーム駆動型研究

Background: Colorectal cancer (CRC) has emerged as the third most prevalent malignancy worldwide. The pursuit of dependable molecular signatures stands as a crucial endeavor for tailoring treatment approaches, refining prognostic assessments, and heightening therapeutic efficacy in CRC management. This investigation was conducted to elucidate essential genes and molecular mechanisms associated with ferroptosis in CRC through implementing machine-learning approaches and AUcell analysis. Methods: The GEO repository and FerrDb served as primary sources for extracting information of gene sets on colorectal cancer and iron-dependent cell death mechanisms. To determine potential therapeutic targets with biomarker significance, we implemented LASSO and SVM-RFE methodology. The immune infiltrates were characterized followed by a competing endogenous RNA network analysis. The AUCell R package was utilized to examine the targeted gene activity patterns within individual cell lines using single-cell transcriptome data. The qRT-PCR and Human Protein Atlas (THPA) database were used to validate the expression of target genes. Potential therapeutic were explored through the DGIdb database. Results: Through the application of machine learning methodologies, five genes were identified as pivotal biomarker candidates: AQP8, NOX4, NR5A2, SCD, and TIMP1. The result of AUcell algorithm showed that the distribution of AUC values exhibited a bimodal pattern, with 2733 cells demonstrating elevated AUC values above the threshold of 0.091. The result of qRT-PCR showed that NOX4, SCD, and TIMP1 were significantly upregulated, while the expression of AQP8 and NR5A2 did not exhibit the expected differences. Both mRNA and IHC analyses from HPA database confirmed the abnormal expression of these pivotal candidate biomarkers. Algorithmic assessment via CIBERSORT methodology revealed notable shifts in immune cell composition within the tumor microenvironment of individuals diagnosed with CRC. Furthermore, A competing endogenous RNA network and 51 potential drug candidates were identified. Conclusion: A systematic framework implementing machine-learning approaches and AUcell analysis was established for identifying core ferroptosis genes and validating their functional link to ferroptosis. Meanwhile, a reliable ferroptosis-associated signature was established, which shed new light on the ferroptosis-mediated molecular mechanisms and therapeutic potential underlying CRC.