Integration of single-cell sequencing and bulk transcriptome data develops prognostic markers based on PCLAF stem-like tumor cells using artificial neural network in gastric cancer.

Background: Gastric cancer stem cells (GCSCs) are important tumour cells involved in tumourigenesis and gastric cancer development. However, their clinical value remains unclear due to the limitations of the available technologies. This study aims to explore the clinical significance of GCSCs, their connection to the tumour microenvironment, and their underlying molecular mechanisms.

Methods: Stem-like tumour cells were identified by mining single-cell transcriptomic data from multiple samples. Integrated analysis of single-cell and bulk transcriptome data was performed to analyse the role of stem-like tumour cells in predicting clinical outcomes by introducing the intermediate variable mRNA stemness degree (SD). Consensus clustering analysis was performed to develop an SD-related molecular classification strategy to assess the clinical characteristics in gastric cancer. A prognostic model was constructed using a customized approach that comprehensively considered SD-related gene signatures based on an artificial neural network.

Results: By analysing single-cell data and validating immunofluorescence results, we identified a PCLAF stem-like tumour cell population in GC. By calculating SD, we observed that PCLAF stem-like tumour cells were associated with poor prognosis and certain clinical features. The SD was negatively correlated with the abundance of most immune cell types. Furthermore, we proposed an SD-related classification method and prognostic model. In addition, the customised prognostic model can be used to predict whether a patient respond to PD-1/PD-L1 immunotherapy.

Conclusion: We identified a cluster of stem-like cells and elucidated their clinical significance, highlighting the possibility of their use as immunotherapeutic targets.