Application of the Pollutants-FCNN Framework:A Multi-Task Neural Network Approach for Con-taminant Toxicity Prediction Based on Tox21Data
Jinan University, 511443 China
Guangzhou Institute of Geochemistry, University of Chinese Academy of Sciences, 510640 China
Jinan University, 511443 China
Abstract
Contaminant toxicity prediction is a pivotal technology for environmental risk assessment and chemical safety management, but traditional experimental methods are limited by high resource consumption, prolonged test cycles, and ethical controversies related to animal experimentation. This study proposes the Pollutants-FCNN Framework, a multi-task learning-based neural network architecture integrated with a multi-head attention mechanism, dedicated to predicting four major toxicity types (biological toxicity, cell toxicity, neurotoxicity, genotoxicity) using the Tox21 benchmark dataset. The framework processes 801 dimensional chemical feature vectors through a standardized pipeline including data preprocessing, attention-augmented FCNN modeling, and multi-task optimization. All core mathematical formulations are derived from the framework’s implementation code to ensure consistency between theoretical design and practical application. Experimental validation shows that the Pollutants-FCNN Framework outperforms conventional pure FCNN models and single-task learning paradigms, achieving a macro-average Area Under the ROC Curve (AUC) of 0.789. This method provides an efficient and reliable in silico toxicity assessment tool, resolving key challenges such as data scarcity for specific endpoints and inefficient feature interaction capture in toxicological modeling. The framework’s modular design and reproducible protocol make it suitable for application in environmental risk assessment and chemical safety screening.
1. Integrates multi-task learning and multi-head attention mechanism to address core challenges of data scarcity and inefficient feature interaction capture in toxicological modeling.
2. Achieves a macro-average AUC of 0.789 on the Tox21 dataset, outperforming traditional pure FCNN and single-task learning models.
3. Features modular design and reproducible protocol, which is applicable to environmental risk assessment and chemical safety screening scenarios.
Keywords:
Contaminant Toxicity Prediction; Multi-Task Neural Network; Multi-Head Attention; Feedforward Neural Network(FCNN); Tox21 DatasetData Availability Statement
The data used in this study are derived from the publicly available Tox21 dataset released by the U.S. Environmental Protection Agency (EPA) and the National Institutes of Health (NIH). Additional data and code supporting the findings of this study are available from the corresponding author upon reasonable request.
Copyright Notice & License:
All articles published in Climate Sustainability & Global Systems (CSGS) are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
Under this license, authors retain copyright and grant the publisher a non-exclusive license to publish the work. Users are free to share, copy, distribute, and adapt the material for any purpose, including commercial use, provided appropriate credit is given to the original author(s) and source, a link to the license is provided, and any changes made are indicated.
The full license terms are available at:
https://creativecommons.org/licenses/by/4.0/
This work is licensed under a Creative Commons Attribution 4.0 International License.
PDF Downloads
Journal Info
How to Cite
References
Ba, J.L., Kiros, J.R., Hinton, G.E., 2016. Layer normalization. arXiv preprint arXiv:1607.06450.
Bishop, C.M., 2006. Pattern recognition and machine learning. Springer.
Chen, J., Si, Y.W., Un, C.W., Siu, S.W., 2021. Chemical toxicity prediction based on semi-supervised learning and graph convolutional neural network. Journal of Cheminformatics 13, 70.
Eastman, P., Pande, V.S., 2019. Predicting toxicity from gene expression with neural networks.
Egan, J.P., 1975. Signal detection theory and roc analysis. Psychology Press.
Gray, M., Wu, L., 2025. Comparative study of molecular descriptors and ai-based embeddings for toxicity prediction. Chemical Research in Toxicology.
Hao, Y., Duan, Z., Liu, L., et al., 2025. Development of an interpretable machine learning model for neurotoxicity prediction of environmentally related compounds. Environmental Science & Technology.
Hastie, T., Tibshirani, R., Friedman, J.H., 2009. The elements of statistical learning. Springer.
Haykin, S., 1994. Neural networks: A comprehensive foundation. Prentice Hall.
He, K., Zhang, X., Ren, S., Sun, J., 2016. Deep residual learning for image recognition, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778.
Huang, R., Xia, M., Nguyen, D.T., et al., 2016. Tox21 challenge to build predictive models of nuclear receptor and stress response pathways as mediated by exposure to environmental chemicals and drugs. Frontiers in Environmental Science 3, 85.
Igarashi, Y., Nakatsu, N., Yamashita, T., et al., 2015. Open tg-gates: a large-scale toxicogenomics database. Toxicology and Applied Pharmacology 289, 534–543.
Ioffe, S., Szegedy, C., 2015. Batch normalization: Accelerating deep network training by reducing internal covariate shift, in: Proceedings of the 32nd International Conference on Machine Learning (ICML), pp. 448–456.
Kim, S., Tariq, S., Heo, S., Yoo, C., 2024. Interpretable attention-based multi-encoder transformer based qspr model for assessing toxicity and environmental impact of chemicals. Chemosphere.
Kingma, D.P., Ba, J., 2015. Adam: A method for stochastic optimization, in: Proceedings of the 3rd International Conference on Learning Representations (ICLR).
Limbu, S., Zakka, C., Dakshanamurthy, S., 2022. Predicting dose-range chemical toxicity using novel hybrid deep machine-learning method. Toxics 10, 706.
Mayr, A., Klambauer, G., Unterthiner, T., Hochreiter, S., 2016. Deeptox: Toxicity prediction using deep learning. Frontiers in Environmental Science 3, 80.
Okey, R., Martis, M., 1999. Molecular level studies on the origin of toxicity: Identification of key variables and selection of descriptors. Chemosphere.
Pantic, I., Paunovic, J., Cumic, J., et al., 2022. Artificial neural networks in contemporary toxicology research. Chemico-Biological Interactions.
Pérez-Santín, E., de-la Fuente-Valentín, L., et al., 2023. Applicability domains of neural networks for toxicity prediction. AIMS Mathematics.
van Rijsbergen, C., 1979. Information retrieval. Butterworth-Heinemann.
Servien, R., Latrille, E., Patureau, D., Hélias, A., 2021. Machine learning models based on molecular descriptors to predict human and environmental toxicological factors in continental freshwater.
Shoji, R., Kawakami, M., 2006. Prediction of genotoxicity of various environmental pollutants by artificial neural network simulation. Molecular Diversity.
Srivastava, N., Hinton, G.E., Krizhevsky, A., et al., 2014. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research 15, 1929–1958.
Vaswani, A., Shazeer, N., Parmar, N., et al., 2017. Attention is all you need, in: Proceedings of the 31st Conference on Neural Information Processing Systems (NeurIPS), pp. 5998–6008.
