scRNA-seq-derived deconvolution and prognostic risk model for lung cancer

Abstract

Aim: Lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC), the two major subtypes of non-small cell lung cancer (NSCLC), possess different immune profiles and potential clinical features. The complex and heterogeneous nature of the tumor microenvironment (TME) demands cell-type–resolved transcriptomic modeling to overcome current limitations in prognostic prediction and therapeutic decision-making.

Methods: Through a comprehensive transcriptomic analysis of publicly available single-cell RNAseq datasets, we associated certain immune cell types with prognostic features. We then ranked cell-type-specific marker genes and created prognostic risk models for each lung cancer subtype using a univariate Cox regression approach. We investigated the prognostic potential of our risk score through Kaplan–Meier analysis for overall survival and validated it with external cohorts.

Results: We have created disease subtype-specific reduced models with shared genes (such as GZMB, DUSP4, FCER1G, C1QA/B, and IRF7), which also performed comparably well.

Conclusions: This study introduces a unique approach to developing prognostic risk scores by comprehensively integrating multiomic data modalities. These models can be utilized in routine clinical monitoring stages in a personalized manner and can help to reduce the burden on healthcare practices.