Interpretable Multi-Modality Consensus QSAR Framework: Integrating Machine and Deep Learning for Enhanced Multi-Endpoint Toxicity Assessment

Authors:FAUZAN SYARIF NURSYAFI¹, MUHAMMAD ADNAN PRAMUDITO², YUNENDAH NUR FUADAH³, and KI MOO LIM¹,⁴,⁵** ¹ Computational Medicine Lab, Department of Medical IT Convergence Engineering, Kumoh National Institute of Technology, Gumi, 39177, Republic of Korea² Computational Medicine Lab, Department of IT Convergence Engineering, Kumoh National Institute of Technology, Gumi, 39177, Republic of Korea³ Telecommunication Engineering Study Program, School of Electrical Engineering, Telkom University Main Campus, Bandung, Indonesia⁴ Computational Medicine Lab, Department of Biomedical Engineering, Kumoh National Institute of Technology, Gumi, 39177, Republic of Korea⁵ Meta Heart Co., Ltd, Gumi, 39253, Republic of Korea Corresponding authors: [e-mail] / [e-mail] 🧩 Overview This repository provides curated chemical datasets for eight mechanistically diverse toxicity endpoints, including standardized training, test, and external validation splits with accompanying SMILES and toxicity labels. It also includes processed outputs from applicability domain (AD) assessment and SHAP-based explainable AI analyses, enabling transparent evaluation of model reliability, interpretability, and chemical space coverage across individual and consensus QSAR models. For users interested in reproducing or extending the full model development workflow—including descriptor generation, machine learning and deep learning model training, consensus construction, and detailed performance evaluation—the complete implementation and notebooks are available in the associated GitHub repository: https://github.com/kit-cml/QSAR-consensus-framework-study. 📚 Citation Citation details will be updated upon publication. 🧠 Acknowledgments This work was conducted at the Computational Medicine Lab, Kumoh National Institute of Technology, Gumi, Republic of Korea.