Feed

Research on molecular classification of lung cancer based on transcriptomic features has achieved remarkable progress. The complementary information provided by distinct molecular profiles has motivated the integration of multi-omics datasets to refine the classification system for lung cancer. In this study, we employed a computational pipeline incorporating 10 clustering algorithms to integrate multi-omics datasets from lung adenocarcinoma (LUAD) patients, combined with 10 machine learning methods, leading to the identification of high-resolution molecular subtypes and the development of a consensus machine learning-driven signature (CMLS) with robust predictive performance. Our findings reveal that the CS1 subtype is associated with more favorable prognosis and enhanced immune responsiveness. Furthermore, a CMLS model constructed from 26 core genes demonstrated strong prognostic predictive power. Patients with high CMLS scores exhibited lower infiltration of CD8⁺ T cells, poorer survival, and diminished response to immunotherapy. In contrast, the low-CMLS group showed improved clinical outcomes, greater responsiveness to immunotherapy, and a tendency toward an immunologically "hot" tumor phenotype. Integrated multi-omics analysis indicated that Gremlin-1 (GREM1) acts as a key regulator within the differential screening-selected gene aberrant in neuroblastoma (DAN) family genes-mediated transforming growth factor-beta (TGF-β) signaling pathway. In conclusion, our data establish a molecular classifier that stratifies patients into distinct score groups, with those in the low-CMLS group potentially benefiting from treatment with pilaralisib.