A two-phase comprehensive NSCLC prognostic study identifies lncRNAs with significant main effect and interaction

Long noncoding RNA (lncRNA) are involved in regulating physiological behaviors for various malignant tumors, including non-small-cell lung cancer (NSCLC). However, few studies comprehensively evaluated both lncRNA–lncRNA interaction effects and main effects of lncRNA on overall survival of NSCLC. Hence, we performed a two-phase designed study of lncRNA expression in tumor tissues using 604 NSCLC patients from The Cancer Genome Atlas as the discovery phase and 839 patients from Gene Expression Omnibus as the validation phase. In the discovery phase, we adopted a two-step strategy, Screening before Testing, for dimension reduction and signal detection. These candidate lncRNAs first screened out by the weighted random forest (Ranger), were then tested through the Cox proportional hazards model adjusted for covariates. Significant lncRNAs with either type of effects aforementioned were carried forward into the validation phase to confirm their significances again. As a result, in the discovery phase, 19 lncRNAs were identified by Ranger, among which five lncRNAs and one pair of lncRNA–lncRNA interaction exhibited significant effects (FDR-q ≤ 0.05) main and interaction effects on NSCLC survival, respectively, through Cox model. After the independent validation, we finally observed that one lncRNA (ENSG00000227403.1) with main effect was robustly associated with NSCLC prognosis (HR(discovery) = 0.90, P = 1.20 × 10(–3); HR(validation) = 0.94, P = 4.11 × 10(–3)) and one pair of lncRNAs (ENSG00000267121.4 and ENSG00000272369.1) had significant interaction effect on NSCLC survival (HR(discovery) = 1.12, P = 3.07 × 10(–4); HR(validation) = 1.11, P = 0.0397). Our comprehensive NSCLC prognostic study of lncRNA provided population-level evidence for further functional study. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s00438-022-01869-3.