Interpretable Prediction Of SARS-CoV-2 Drug Efficacy And Cytotoxicity Using Multivariate Adaptive Regression Splines
Abstract
Accurate prediction of drug efficacy and cytotoxicity for SARS-CoV-2 is a critical step in early-stage drug development, guiding compound prioritization and identifying potential therapies. Ensemble-based algorithms such as Random Forest have demonstrated strong predictive performance in this domain; however, their reliance on post hoc interpretability methods like SHAP (Shapley Additive Explanations) often yields complex, global explanations of feature importance rather than simple, actionable rules. To enhance direct interpretability while maintaining competitive predictive accuracy, this study applies Multivariate Adaptive Regression Splines (MARS), a nonparametric modeling framework that provides rule-based transparency, to a carefully prepared SARS-CoV-2 dataset integrating network and physicochemical features. Comparative receiver operating characteristic (ROC) analyses demonstrated that MARS achieved consistently strong discriminative ability in cytotoxicity-focused models, while showing relatively moderate performance for efficacy-related classification. These findings reveal a trade-off between interpretability and predictive performance, suggesting that while simpler, transparent models can effectively capture determinants of cytotoxicity, they may generalize less efficiently for efficacy prediction. Overall, this study highlights the potential of intrinsically interpretable machine learning frameworks such as MARS for generating clear, mechanistic insights into drug safety profiles.
Keywords
Machine learning, MARS, Nonparametric modeling, ROC analysis, SARS-CoV-2
References
- Gordon, D. E., Jang, G. M., Bouhaddou, M., et al. (2020). A SARS-CoV-2 protein interaction map reveals targets for drug repurposing. Nature, 583(7816), 459–468. https://doi.org/10.1038/s41586-020-2286-9
- Hay, M., Thomas, D. W., Craighead, J. L., et al. (2014). Clinical development success rates for investigational drugs. Nature Biotechnology, 32(1), 40–51. https://doi.org/10.1038/nbt.2786
- Scannell, J. W., Blanckley, A., Boldon, H., et al. (2012). Diagnosing the decline in pharmaceutical R&D efficiency. Nature Reviews Drug Discovery, 11(3), 191–200. https://doi.org/10.1038/nrd3681
- Späth, J., Wang, R. S., Humphrey, M., et al. (2024). Machine learning-based integration of network features and chemical structure of compounds for SARS-CoV-2 drug effect analysis. CPT: Pharmacometrics & Systems Pharmacology, 13(2), 257–269. https://doi.org/10.1002/psp4.13076
- Rodriguez, S., Hug, C., Todorov, P., et al. (2021). Machine learning identifies candidates for drug repurposing in Alzheimer's disease. Nature Communications, 12(1), 1033. https://doi.org/10.1038/s41467-021-21330-0
- Rapicavoli, R. V., Alaimo, S., Ferro, A., et al. (2022). Computational methods for drug repurposing. Advances in Experimental Medicine and Biology, 1361, 119–141. https://doi.org/10.1007/978-3-030-91836-1_7
- Meyer, J. G., Liu, S., Miller, I. J., et al. (2019). Learning drug functions from chemical structures with convolutional neural networks and random forests. Journal of Chemical Information and Modeling, 59(10), 4438–4449. https://doi.org/10.1021/acs.jcim.9b00236
- Zhu, J., Wang, J., Wang, X., et al. (2021). Prediction of drug efficacy from transcriptional profiles with deep learning. Nature Biotechnology, 39(11), 1444–1452. https://doi.org/10.1038/s41587-021-00946-z
- Lundberg, S. M., & Lee, S.-I. (2017). A unified approach to interpreting model predictions. In I. Guyon et al. (Eds.), Advances in Neural Information Processing Systems (Vol. 30, pp. 4765–4774). Curran Associates, Inc.
- Friedman, J. H. (1991). Multivariate adaptive regression splines. The Annals of Statistics, 19(1), 1–67. https://doi.org/10.1214/aos/1176347963