Structural Classification of PFAS using Molecular Fingerprints and Graph Networks

This project provides a network-based structural classification resource for 13,028 per- and polyfluoroalkyl substances (PFAS) from the U.S. EPA’s 2024 PFAS8a7v3 list. Using eight molecular fingerprinting methods and K-Nearest Neighbor Graphs (K-NNGs), we assign Proximity Classes to 288 previously unclassified PFAS compounds. The dataset includes Proximity Class outputs, UMAP coordinates, edge lists, and interactive 3D network visualizations across multiple neighborhood sizes. All files are designed for reuse in regulatory prioritization, chemical substitution, and machine learning applications.

本项目针对美国环境保护署（U.S. EPA）2024年PFAS8a7v3清单中的13028种全氟和多氟烷基物质（per- and polyfluoroalkyl substances, PFAS），提供了基于网络的结构分类资源。本研究采用8种分子指纹表征方法与K近邻图（K-Nearest Neighbor Graphs, K-NNGs），为288种此前未分类的PFAS化合物分配了邻近类别（Proximity Class）。本数据集包含邻近类别输出结果、均匀流形近似与投影（UMAP）坐标、边列表，以及多邻域尺寸下的交互式三维网络可视化文件。所有文件均设计为可复用于监管优先级排序、化学品替代及机器学习场景。