Systematic Construction of Relative Expression Ordering Atlas for Pan-Cancer Drug Resistance Prediction

Gene relative expression ordering (REO) patterns within samples have been associated with drug responses in specific cancer types, yet a systematic pan-cancer assessment remains lacking. In this study, we integrated transcriptomic data from the Cancer Cell Line Encyclopedia (CCLE) and drug sensitivity profiles from the Genomics of Drug Sensitivity in Cancer (GDSC) to construct a comprehensive pan-cancer REO signature atlas for dissecting tumor drug resistance mechanisms. We systematically analyzed 1,923 distinct drug–cancer conditions, each comprising at least 10 sensitive and 10 resistant cell lines. Differentially expressed genes were converted into REO gene-pair matrices, and robust features were preserved via Fisher’s exact test and stable reversal proportion (SRP) filtering. A drug resistance prediction model was subsequently established using cross-validation, recursive feature elimination (RFE), and the random forest algorithm. REO features demonstrated strong stability, high discriminative capacity, and exceptional cross-platform robustness. The REO-based framework achieved outstanding predictive performance, with accuracy ≥ 0.80 in 96% of conditions and AUC ≥ 0.80 in 93.5% of conditions, significantly surpassing previously reported REO-based approaches. Internal validation using the Cancer Therapeutics Response Portal (CTRP) confirmed significant correlations between model predictions and empirical drug sensitivity (P < 0.05). Pan-cancer analysis identified 29 recurrent core REO gene pairs that were consistently enriched in key drug resistance pathways, including extracellular matrix remodeling, cytokine signaling, and transmembrane potential regulation. Further external validation with GEO datasets and single-cell RNA-seq (GSE156246) verified that REO features reliably capture early transcriptional reprogramming and dynamically monitor the enrichment of drug-resistant clones during treatment. Collectively, this pan-cancer REO atlas serves as an interpretable, stable, and transferable molecular resource for elucidating conserved drug resistance mechanisms, identifying therapeutic targets, and facilitating companion diagnostic development in precision oncology.