CADRE CAncer Drug Response nEtworks

Amid the growing challenges of drug development, the prediction of synergistic drug combinations in cancer therapy emerges as a critical problem with potentially profound clinical implications. However, precise forecasting of synergy remains elusive due to the complex biological determinants influencing cell responsiveness to therapies. Drug response data sets, encompassing single-agent and combinatorial treatments applied to cancer cell lines, have commonly served as the foundation for training and validating synergy prediction models. Intriguingly, these models, based on both small and large-scale data sets, have reported successes despite the inherent limitations of using IC50 values for response prediction and potential overfitting. This study introduces CADRE (CAncer Drug Response nEtwork), a conceptually simple and relatively low-complexity computational model for drug synergy prediction. By analyzing how similar cancer cell lines respond to individual drugs, CADRE predicts the combined effect of drug combinations. Remarkably, CADRE frequently outperforms state-of-the-art machine learning algorithms across diverse cancer panels. We further explore the intrinsic suitability of these data sets for synergy prediction and the generalizability of trained models to more complex cancer systems, emphasizing the critical role of understanding underlying biological mechanisms. Importantly, our and previously established models exhibit reduced performance when applied to data sets from more heterogeneous cancer panels. We delve into the unique characteristics of melanoma as a study subject compared to other cancer cohorts with similar data, offering a plausible biological explanation for this discrepancy.

Availability: Source code and supplementary data are available at CADRE github repo.

Presented at: RECOMB CCB 2024

Read the full paper: [coming soon](Lai et al., 2024)