PFAS Toxicokinetics with Gnns: Predicting Half-Life for Risk and Remediation

Overview

PFAS are synthetic chemicals found in various species and environments. Due to the vast number of PFAS compounds and research constraints, information about their half-lives is limited. This study aims to predict the half-lives of PFAS using a novel graph neural network (GNN) model, using available data extrapolated from various studies. The GNN analyzes the chemical structure of PFAS compounds along with graph-level characteristics, including species, sex, and dosing method of PFAS contamination. Chemical structures were obtained by converting the CASRN identifiers to SMILES codes and then to graph representations using PyTorch. Additional features extracted for each atom include atomic number, degree, formal charge, number of radical electrons, chiral tag, aromaticity, and ring membership. Bond features such as type, conjugation, ring membership, and direction were also incorporated. This model is novel because it uses the molecular structure of 6,472 different PFAS compounds to predict half-life. So, if a new compound is developed, this model will perform without additional modification. The training data was grouped to include an equal number of records for each species, and k-means clustering was used to train the model, achieving an accuracy of 82.5%. This model utilizes publicly available data to predict the half-life of PFAS, providing a valuable tool for estimating PFAS decomposition rates in various environments and species. Such predictions can inform risk assessments, guide remediation efforts, and support policy decisions on PFAS management and regulation

Competition history

• AJAS 2026 Category not listed

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• AJAS None projects
• Every AJAS 2026 project