Data archive for aquatic ecotoxicity representation benchmarking

This record provides the processed datasets, feature matrices, scaffold-based split files, graph resources, model prediction outputs, and merged result tables used to support verification and reproducibility of the manuscript-level analyses for “Comparative Evaluation of Molecular Representations for Aquatic Ecotoxicity Prediction: Fingerprints, Molecular Graphs, and Similarity Graphs”. The archive includes processed TEST-derived toxicity tables, Daphnia regression targets, Morgan fingerprint features, RDKit descriptor features, scaffold-based split files, the TopK-50 G2 similarity graph edge table, model-level test predictions, inference-target predictions, and merged prediction tables. The source code used for data processing, model training, inference, and downstream analysis is available at: https://github.com/rayah0123/aquatic-toxicity-representations.本数据集包含经预处理的各类数据集、特征矩阵、基于骨架的划分文件（scaffold-based split files）、图谱资源、模型预测输出与合并结果表，旨在支撑《用于水生生态毒性预测的分子表征对比评估：指纹图谱、分子图谱与相似性图谱》一文的手稿级分析的验证与可重复性研究。 本归档文件涵盖经预处理的TEST衍生毒性表、水蚤回归靶标数据、摩根指纹（Morgan fingerprint）特征、RDKit描述符特征、基于骨架的划分文件、TopK-50 G2相似性图谱边表、模型级测试预测结果、推理靶标预测结果与合并预测表。 本研究用于数据预处理、模型训练、推理及下游分析的源代码可通过以下链接获取：https://github.com/rayah0123/aquatic-toxicity-representations。