Abstract
A comprehensive analysis of the transcriptomic landscape in early-stage LUAD (stages 1 and 2) can identify robust survival and staging predictors. Using a systems-biology approach with TCGA LUAD transcriptomic data, we identified 18 co-expression modules, 11 correlated with staging and 7 with survival. Five survival- and staging-associated (SAS) modules (M1, M3, M6, M9, and M16) distinguished patients with differing survival curves based on module eigengene expression. Anti-correlated SAS-modules (M3 and M6) exhibited the strongest associations with OS and staging. Key differentially-expressed hub genes from M3 and M6 were nominated as prognostic markers through iterative combinatorial ROC analysis. Survival prediction was refined by testing equal-weight gene ratios with opposing correlations to survival, revealing additive or synergistic predictive value. The top-performing ratio, ATP6V0E1 + SVBP + HSDL1 + UBTD1/GNPNAT1 + XRCC2 + TFAP2A + PPP1R13L, achieved an average AUC of 75.5% across three timepoints (12 months, 18 months, and 3 years). Comparison with three established LUAD prognostic signatures (Shedden, Soltis, and Song) within the TCGA network showed that our 8-gene signature had comparable or superior predictive power while maintaining concordance with existing signatures, some of which contain an order of magnitude more transcripts. Our findings suggest that novel gene signature holds promise for refining early-stage LUAD prognostic modeling and informing treatment strategies.