Nicoleta Spînu holds a PhD in computational toxicology from Liverpool John Moores University, where her research advanced the concept of quantitative Adverse Outcome Pathways. She is the founder of a start-up specialised in chemical safety assessment for the personal care industry, applying cutting-edge computational methods and multi-modal data integration. Her work translates complex toxicological science into practical solutions that support both product innovation and regulatory compliance.
FL-CHEMSAFE: Advancing Safety Assessment of Chemicals Through Federated Learning
Federated learning has emerged as a promising approach to drug discovery, with real-world implementations such as MELLODY [1] and Effiris [2]. Yet, its application to chemical safety assessment remains unexplored. Data is often siloed, proprietary, or costly to generate. This presentation will discuss how federated learning approaches can overcome data fragmentation in chemical safety assessment. Three use cases were simulated with the Flower open-source federated learning framework, namely (i) federated analytics for dermal permeability (log Kp) screening; (ii) federated convolutional neural networks (CNNs) for mutagenicity prediction from SMILES strings, and (iii) federated eXtreme Gradient Boosting (XGBoost) models for predicting skin sensitisation potential using molecular fingerprints and descriptors. The results show that federated learning can achieve predictive performance comparable to centralised models while expanding the applicability domain, raising important considerations around model governance. By facilitating model development across distributed, proprietary datasets without compromising confidentiality, federated learning can enhance Next-Generation Risk Assessment through more comprehensive and accurate chemical safety assessment.
Acknowledgements: The funding for the feasibility project FL-CHEMSAFE by the MKB innovatiestimulering topsectoren (MIT) Noord-Holland and the Flower Pilot Program are gratefully acknowledged.
References:
[1] Heyndrickx W., et. al. J Chem Inf Model, 64(7), 2024, 2331-2344.
[2] Bassani D., et. al. Chem Res Toxicol, 36(9), 2023, 1503-1517.