Harnessing machine learning-driven multiomics integration: deciphering programmed cell death networks for prognostication and immunotherapy prediction in lung adenocarcinoma.

BACKGROUND: Programmed cell death (PCD) patterns play important roles in lung adenocarcinoma (LUAD) development as well as treatment resistance, and in-depth study of PCD is beneficial for improving the therapeutic paradigm for LUAD. METHODS: Fourteen PCD-related patterns were integrated and multiple datasets from TCGA and GEO were collected to develop a PCD signature using 101 machine learning algorithm combinations. Prognosis, immune cell infiltration, and sensitivity to chemotherapy and immunotherapy were compared between different risk groups and validated by multiple bulk RNA-seq and scRNA-seq datasets of patients receiving immunotherapy. CellChat was used to analyze the cellular interactions between patients with different PCD groups. Immune cell infiltration in the tumor tissues of 38 LUAD patients treated with anti-PD-1 therapy was validated by multiplex immunohistochemistry (mIHC). RESULTS: A PCD signature containing 7 genes was constructed using 101 machine learning algorithm combinations and validated across multiple datasets. High PCD scores in patients are associated with poorer prognosis, lower immune cell infiltration, and reduced responsiveness to immunotherapy. In addition, the PCD signature were comprehensively analyzed by scRNA-seq, and the results showed that the high PCD signature was concentrated mainly in advanced LUAD. Moreover, pathways associated with tumor progression and immune resistance were more strongly promoted in the high PCD signature group. The expression of the key gene NAPSA correlated with immune cell infiltration and immunotherapy response, as confirmed by IHC and mIHC. CONCLUSION: The PCD signature confers significant potential to predict prognosis of LUAD in patients, and NAPSA is promising as a new marker for predicting the efficacy of immunotherapy.