Integrating cell interaction with transcript factors to obtain a robust gene panel for prognostic prediction and therapies in cholangiocarcinoma

Abstract Background The efficacy of immunotherapy in cholangiocarcinoma (CCA) is blocked by its high degree of tumor heterogeneity. Methods We constructed empirical Bayes and Markov random field models to calculate transcription factors, interaction genes, and associated signaling pathways involved in cell-cell communication using single cell RNAseq data. Then, we analyzed the immune exhaustion mechanism in the CCA progression. Results We found that VEGFA positive macrophages with the high level of LGALS9 could interact with HAVCR2 to promote exhaustion of CD8 T cells in CCA. Transcript factors SPI1 and IRF1 could up-regulate the expression of LGALS9 in VEGFA positive macrophages. Subsequently, we obtained a panel containing 54 genes through the model, which could identify the subtype S2 with high expression of immune checkpoint genes that are suitable for immunotherapy. Moreover, we found that patients in subtype S2 with a higher mutation ratio of MUC16 accompanied with immune exhausted genes such as HAVCR2 and TIGIT. Finally, we constructed a nine-gene eLBP-LASSO-COX risk model which was designated as tumor microenvironment risk score (TMRS). Conclusions eLBP could simultaneously analyze the interaction genes, pathways and transcription factors involved in the cell communication. The TMRS panel was revealed to be a reliable tool for prognostic prediction and chemotherapeutic decision-making in cholangiocarcinoma.