Soil pollution is a critical environmental challenge: the substances released in the soil can adversely affect humans and the ecosystem. Several bioassays were developed to investigate the soil ecotoxicity of chemicals with soil microbes, plants, invertebrates and vertebrates. The 28-day collembolan reproduction test with the springtail Folsomia candida is a recently introduced bioassay described by OECD guideline 232. Although the importance of springtails for maintaining soil quality, toxicity data for Collembola are still limited. We have developed two QSAR models for the prediction of reproductive toxicity induced by organic compounds in Folsomia candida using 28 days NOEC data. We assembled a dataset with the highest number of compounds available so far: 54 compounds were collected from publicly available sources, including plant protection products, reactive intermediates and industrial chemicals, household and cosmetic ingredients, drugs, environmental transformation products and polycyclic aromatic hydrocarbons. The models were developed using partial least squares regression (PLS) and the Monte Carlo technique with respectively the open source tools Small Dataset Modeler and CORAL software. Both QSAR models gave good predictive performance even though based on a small dataset, so they could serve for the ecological risk assessment of chemicals for terrestrial organisms.

土壤污染是一项严峻的环境挑战：土壤中释放的物质会对人类和生态系统产生不利影响。开发了几种生物测定法来研究化学品对土壤微生物、植物、无脊椎动物和脊椎动物的土壤生态毒性。使用跳蚤 Folsomia candida进行的 28 天跳虫繁殖测试是最近引入的经合组织指南 232 描述的生物测定法。尽管跳虫对保持土壤质量的重要性，但跳虫的毒性数据仍然有限。我们开发了两个 QSAR 模型，用于预测念珠菌中有机化合物引起的生殖毒性使用 28 天的 NOEC 数据。我们收集了迄今为止可用化合物数量最多的数据集：从公开来源收集了 54 种化合物，包括植物保护产品、活性中间体和工业化学品、家用和化妆品成分、药物、环境改造产品和多环芳烃。这些模型是使用偏最小二乘回归 (PLS) 和 Monte Carlo 技术分别使用开源工具 Small Dataset Modeler 和 CORAL 软件开发的。尽管基于一个小数据集，这两个 QSAR 模型都提供了良好的预测性能，因此它们可以用于陆地生物化学品的生态风险评估。