CoDER: Consistent Drug Efficacy Ranking

Drug ranking plays a pivotal role in drug repurposing and precision medicine by prioritizing candidate therapies under biological and clinical constraints. A major challenge is identifying drugs whose efficacy remains consistent across multiple human tissues, as heterogeneity in gene expression, regulatory mechanisms, and tissue-specific biology can lead to variable or unreliable therapeutic effects. Many existing approaches focus on single tissues or aggregate performance across tissues, obscuring important sources of variability. We present CoDER (Consistent Drug Efficacy Ranking), a graph-based computational framework for identifying drug orderings that remain stable across a minimum number of tissues. CoDER integrates multiple biological data sources-including GTEx, TRRUST, DisGeNET, and DGIdb-to derive tissue-specific drug efficacy profiles and represent pairwise drug dominance relationships as a tissue-labeled directed graph. We formalize this task as a $\lambda$-consistent ordering problem and show that it is NP-hard. To address this computational challenge, CoDER employs a scalable divide-and-conquer strategy that limits exhaustive search to tractable subproblems and iteratively refines candidate sets to recover long $\lambda$-consistent drug sequences. Applied to 273 drugs across 49 human tissues, CoDER efficiently identifies interpretable tissue-consistent rankings and achieves higher scalability and longer $\lambda$-consistent paths than the evaluated tissue-aware ranking baselines. These results suggest that CoDER provides a practical computational framework for tissue-consistent drug prioritization and hypothesis generation.