The production of nanomaterials continues its rapid growth; however, newly manufactured nanomaterials' environmental and health safety are among the most significant concerns. A safety assessment is usually a lengthy and costly process, so computational studies are often used to complement experimental testing. One of the most time-efficient techniques is structure-activity relationships (SAR) modeling. In this project, we analyzed the Sustainable Nanotechnology (S2NANO) dataset that contains 574 experimental cell viability and toxicity datapoints for Al₂O₃, CuO, Fe₂O₃, Fe₃O₄, SiO₂, TiO₂, and ZnO measured in different conditions. We aimed to develop classification- and regression-based structure-activity relationship models using quasi-SMILES molecular representation. Introduced quasi-SMILES took into consideration all available information, including structural features of nanoparticles (molecular structure, core size, etc.) and related experimental parameters (cell line, dose, exposure time, assay, hydrodynamic size, surface charge, etc.). Resultant regression models demonstrated sufficient predictive power, while classification models demonstrated higher accuracy.

纳米材料的生产继续快速增长；然而，新制造的纳米材料的环境和健康安全是最重要的问题之一。安全评估通常是一个漫长而昂贵的过程，因此计算研究通常用于补充实验测试。最省时的技术之一是构效关系 (SAR) 建模。在本项目中，我们分析了可持续纳米技术 (S2NANO) 数据集，其中包含 Al ₂ O ₃、CuO、Fe ₂ O ₃、Fe ₃ O ₄、SiO ₂、TiO _{2 的}574 个实验细胞活力和毒性数据点, 和 ZnO 在不同条件下测量。我们旨在使用准 SMILES 分子表示开发基于分类和回归的构效关系模型。引入 quasi-SMILES 考虑了所有可用信息，包括纳米颗粒的结构特征（分子结构、核心尺寸等）和相关实验参数（细胞系、剂量、暴露时间、测定、流体动力学尺寸、表面电荷等）。 . 结果回归模型表现出足够的预测能力，而分类模型表现出更高的准确性。