Identifying the specifically expressed lncRNAs in the tumor microenvironment

Understanding immune cell heterogeneity in the tumor microenvironment (TME) is crucial for studying tumor immunity and improving immunotherapy strategies. Diverse immune cell populations, particularly T cells, exhibit functional heterogeneity and play distinct roles in the TME, and even within the same immune cell lineage, multiple subtypes or states with distinct transcriptional and functional characteristics can coexist across different cancer types and individual patients. Recent single-cell studies have further revealed remarkable heterogeneity among tumor-infiltrating T cells, particularly within the CD8 + T-cell compartment. Since gene expression programs reflect the functional properties of these cellular subtypes, identifying subtype-specific regulatory molecules is essential for understanding their roles in antitumor immunity. As important regulatory molecules in immune-related molecular mechanisms, long non-coding RNAs (lncRNAs), which often exhibit strong cell-type and context specificity, may serve as important regulators of subtype-specific immune programs. Therefore, identifying lncRNAs that are specifically associated with different subtypes of tumor-infiltrating CD8 + T cells may provide important insights into the regulatory mechanisms of immune responses in cancer. However, detecting subtype-associated lncRNAs from single-cell RNA sequencing (scRNA-seq) data remains challenging due to sparse expression patterns and the low abundance of many lncRNAs. In this study, we developed SDSGI (Surprise-Driven Subtype-Specific Gene Identification), a computational method for identifying subtype-specific genes and lncRNAs from scRNA-seq data. This framework integrates two complementary scoring strategies: an expression-based specificity score derived from the \(\tau \) index and a surprise-information-based score that captures statistically informative subtype-enrichment signals. By combining expression specificity with statistical evidence, the method improves the detection of low-abundance but subtype-associated genes, including lncRNAs that may be missed by conventional differential expression analysis. We applied SDSGI to tumor-infiltrating CD8 + T-cell datasets across multiple cancer types. The results show that the method identifies clear subtype-specific gene signatures and biologically meaningful lncRNAs associated with immune cell states. Specifically, 6,899 subtype-specific genes in breast cancer including 497 lncRNAs; 5,035 genes in B-cell lymphoma including 361 lncRNAs; and 11,092 genes in esophageal cancer including 944 lncRNAs have been identified by our method. Functional enrichment, co-expression network analysis, and survival validation further suggested that the identified lncRNAs are associated with immune regulation, subtype-specific functional modules, and cancer prognosis. These results indicate that SDSGI provides a useful framework for discovering subtype-specific lncRNAs in the tumor immune microenvironment.