Characterizing Chemical Toxicity for Life Cycle Assessment Using Machine Learning Models Based on Environmental Footprint – Illustrated importance through a textile case study

To support the integration of emerging chemicals into life cycle assessment (LCA), we developed a machine learning (ML) workflow to predict characterization factors (CFs) for both human toxicity and ecotoxicity, aligned with the EU Environmental Footprint (EF) methodology. Using extended chemicals from EF version 3 database and molecular descripotrs derived from SMILES (Simplified Molecular-Input Line-Entry System) strings, we trained and evaluated three ML models: extreme gradient boosting (XGBoost), gaussian process (GP) regression, and deep neural networks (NN). A clustering step was used to guide model selection for new compounds. XGBoost consistently performed best, achieving R² values up to 0.65 and 0.61 for ecotoxicity and human toxicity (seas water, continent). New chemicals should first be assigned to clusters for which the best tailored ML model is then selected. An LCA case study in the textile sector illustrates the importance of CF prediction, as the total human toxicity score is at least 4 orders of magnitudes higher when including predicted CFs that were originally missing. Overall, this end-to-end ML approach offers a robust and efficient alternative to traditional methods, enabling immediate integration of predicted CFs into LCA studies, which could facilitate safe and sustainable by design implementation.