Contradictory of gene expression signatures for clinical prognostic prediction caused by update of reference genome and annotation. Contradictory of gene expression signatures for clinical prognostic prediction caused by update of reference genome and annotation

Gene expression estimates detected by RNA-sequencing technology vary with the updates of reference genome and gene annotation, which might confound existing expression-based prognostic signatures, making them inapplicable to clinical practice. In this study, we proposed a method to decrease these effects and developed a qualitative signature for stage I lung adenocarcinoma, whose classification was based on within-sample relative expression orderings (REOs) of gene pairs. The signature was validated in 471 stage I samples derived from public RNA-sequencing and microarray data (both log-rank p < 0.001). Notably, our signature could effectively predict prognosis for 30 stage I patients with severely degraded FFPE tissues (log-rank p = 0.0177). More important, the risk classification was stable in the latest annotation. In summary, our signature would be a promising signature for clinical individualized application because of its excellent prognostic performance and classification robustness. Overall design: 30 FFPE tissues with severely RNA degradation from stage I lung adenocarcinoma patients who had surgery with curative intent were sequenced by RNA-Seq.