機械学習手法を用いた多がん診断のための組織および血液中の白血球の転写シグネチャの特定

Investigating the transcriptional signatures of immune cells in various cancer types is crucial for understanding their roles in the tumor microenvironment and developing effective immunotherapeutic strategies. In this study, we employed machine learning methods to analyze RNA-seq data from patients with four different types of cancers and two immune cell types, including T cell and CD45+CD3- leukocyte cell types. We processed seven datasets, each divided into three groups on the basis of cell source: tumor, normal adjacent tissue, and peripheral blood. The datasets were downscaled by using the Boruta method, and the remaining genes were ranked for criticality in a list through the max-relevance and min-redundancy method. The obtained list of genes was fed into incremental feature selection (IFS), which employed decision tree or random forest to distinguish cells, for the identification of key genes associated with immune cell function in different cancer types and construction of efficient classifiers and classification rules (special patterns for different groups). Our results revealed distinct expression patterns of key genes, such as the downregulation of CST7 in T cells from tumor tissues and differential expression of CD2 in non-tumor sites. Furthermore, we identified LCP1, CD27, and MAL as immunologically relevant genes in T cells across different tissue origins, whereas IFI30, CXCR4, and FOSB played various roles in CD45+CD3- leukocytes. The identified key genes were supported by evidence in the literature, highlighting their involvement in antitumor processes in T cells and other immune cells. Our findings provide valuable insights into the transcriptional signatures of immune cells in different cancer types and lay the foundation for the development of novel diagnostic, prognostic, and therapeutic strategies in cancer immunology.