Identification of a novel lncRNA prognostic signature for lung adenocarcinoma via ML-based feature selection and artificial neural networks

Lung cancer remains the leading cause of cancer-related deaths with lung adenocarcinomas as the most common subtype. Emerging evidence suggests that long non-coding RNAs (lncRNAs) are integral to cancer biology and present significant promise as prognostic biomarkers and therapeutic targets. This study aimed to identify a prognostic lncRNA signature for patients with lung adenocarcinoma by employing machine learning techniques on a combined 349-case TCGA-LUAD and OncoSG-LCCS gene expression dataset, classifying patients by longer (> 3 years) versus shorter (< 2 years) overall survival phenotypes. After pre-processing steps, recursive feature elimination with support vector machines and ridge classifier was applied to select relevant lncRNA features, leading to the identification of a novel 40-lncRNA signature associated with overall survival. Survival analysis demonstrated the prognostic value of this 40-lncRNA signature, yielding statistically significant results in both favorable ( p = 0.00018, HR = 0.63) and unfavorable ( p = 0.00011, HR = 1.6) subsets. Enrichment analysis revealed immune- and cell cycle-related pathways associated with these lncRNAs, highlighting their biological relevance in lung adenocarcinoma. Moreover, classification modeling showed improved accuracy with this 40-lncRNA signature compared to the full dataset. Notably, artificial neural networks demonstrated the highest performance (accuracy = 0.805 ± 0.02, AUC = 0.85 ± 0.03). Key features of this model, including DPP10-AS1 and FOXD3-AS1, are recognized for their significant contributions to lung adenocarcinoma progression. This study introduces a novel lncRNA signature with potential applications in personalized prognosis and therapeutic development for lung adenocarcinoma, supporting future research on discovering lncRNA biomarkers.