Genomic-Adjusted Radiation Dose from Bulk RNA Sequencing for Personalized Radiotherapy

Abstract

Radiotherapy is delivered to more than half of all patients with cancer yet is prescribed using uniform physical doses despite well-established interpatient variability in biological response. The genomic-adjusted radiation dose (GARD), derived from the radiosensitivity index (RSI), integrates tumor transcriptomics with radiation dose to estimate patient-specific treatment effect, and has been clinically validated as a predictor of radiotherapy benefit across diverse disease sites, including breast, lung, head and neck, glioma, sarcoma, rectal, and endometrial cancers. However, further clinical validation and deployment has been limited by reliance on microarray-based expression. Here we develop an RNA sequencing-based formulation of RSI (RSI-seq) and show that it preserves the functional properties of the original model across measurement platforms. RSI-seq maintains concordance with microarray RSI, including preservation of patient rank ordering (pooled Spearman{rho} = 0.86), and, when integrated into GARD, reproduces predicted changes in biological effect under clinically relevant dose perturbations (R2 [≥] 0.78 for {Delta}GARD in both directions). This preservation of interventional prediction is robust to expression noise and invariant to normalization strategy, enabling consistent application across RNA-seq pipelines. Application across the TCGA pan-cancer transcriptomic atlas demonstrates scalability across tumor types, with cohort medians agreeing closely with previously published microarray RSI medians (Spearman{rho} = 0.68, Pearson r = 0.85 across 20 matched cohorts). By bringing a clinically validated radiogenomic dose model into the RNA-sequencing era, RSI-seq makes biologically personalized radiotherapy directly accessible, retrospectively in existing RNA-seq cohorts and prospectively in modern clinical sequencing workflows, across the full range of tumor types treated with radiation.