The complex relationship between environmental exposure and human health constitutes a major global challenge requiring innovative solutions. Artificial Intelligence (AI) and Machine Learning (ML) show exceptional strength for data analysis and pattern recognition. Applying these technologies to environmental health provides new insights to improve and advance environmental exposure assessment, health risk assessment, and related policy development. It is with great pleasure that we present this Special Issue of Environment & Health on Machine Learning and Artificial Intelligence for Environmental Health. This collection of research highlights the latest advancements and broad potential of ML and AI to empower our response to pressing and future environmental health issues.
The most common environmental applications of AI/ML are toxicity prediction of certain pollutants. As important toxicity prediction tools, Quantitative Structure–Activity Relationship (QSAR) and Quantitative Structure–Property Relationship (QSPR) models aim to predict compound bioactivity and toxicity based on their structure information. AI/ML has powered the performance of QSPR models. Liu et al. applied both classic ML models and deep learning models to assess if chemicals are lung surfactant inhibitors and multiplayer perception (MLP) model showed the best performance (DOI: 10.1021/envhealth.4c00118). In another study, ensemble learning based AquaticTox, combining six diverse machine and deep learning methods, including GACNN, Random Forest, AdaBoost, Gradient Boosting, Support Vector Machine, and FCNet, was developed to predict aquatic toxicity of organic compounds in five aquatic species: Oncorhynchus mykiss, Pimephales promelas, Daphnia magna, Pseudokirchneriella subcapitata, and Tetrahymena pyriformis and outperformed all single models (DOI: 10.1021/envhealth.4c00014). Furthermore, AquaticTox incorporates a knowledge base of structure-aquatic toxic mode of action (MOA) relationships, shedding lights on the mechanisms underlying chemical toxicity.
Though better prediction performance of QSPR models is important, explainable AI (XAI) can further interpret the machine learning models. XAI helps to understand “black box” models, improving transparency in model predictions, which is essential for their applications in regulatory and public health decision-making. Utilizing the Local Interpretable Model-agnostic Explanations (LIME) method in conjunction with Random Forest (RF) classifier models, Rosa et al. identified molecular fragments impacting five key nuclear receptor targets: androgen receptor (AR), estrogen receptor (ER), aryl hydrocarbon receptor (AhR), aromatase receptor (ARO), and peroxisome proliferator-activated receptors (PPAR) (DOI: 10.1021/envhealth.4c00218).
The lack of experimental data is the bottleneck for the QSPR model. Mixture toxicity is one of those scenarios. In the work by Xu et al., the linear QSAR model was developed to predict time dependent toxicities of binary mixtures of five antibiotics and found the number of hydrogen-bonded donor and positively charged pharmacophore point pairs at a topological distance of four bonds will significantly influence such mixture toxicity (DOI: 10.1021/envhealth.4c00001). With limited data and complex mechanisms, immunotoxicity is also hard to predict. Utilizing large public toxicity data sets, Daood et al. showed a computational modeling strategy to build QSAR models to connect various AhR related key events and reveal a potential immunotoxicity mechanism (DOI: 10.1021/envhealth.4c00026).
Environmental exposure assessment requires high resolution spatial and temporal data, which is usually hard to get with regular environmental monitoring. Machine learning could overcome this issue. For example, different machine learning models can be trained to perform spatial prediction of nationwide daily PM2.5, as well as an ensemble model. Such a method could be used to access short-term health risks (DOI: 10.1021/envhealth.4c00201).
Besides air pollution, AI and ML are also involved in next generation risk assessment. One crucial step is in vitro to in vivo extrapolation (IVIVE), which will communicate bioactive chemical concentrations from in vitro assays to in vivo applications. Han et al. reviewed current IVIVE studies and highlighted both physiologically based toxicokinetic (PBTK) models and machine learning algorithms and their applications in chemical prioritization, hazards assessment, and regulatory decision making (DOI: 10.1021/envhealth.4c00043).
The relationship between environmental factors and health outcomes has also benefited from AI/ML. Midya et al. investigated the influence of metals on intestinal inflammation considering pregnancy exposure and childhood gut microbiome. An interpretable algorithm called the “repeated hold-out signed-iterated Random Forest” (rh-SiRF) helps to identify multiordered combinations of predictors, so-called “metal-microbial clique signatures”, that occur and are associated with health outcome. Results showed two metal-microbial clique signatures and provided a framework for a “precision environmental health” (DOI: 10.1021/envhealth.4c00125).
AI/ML could also improve environmental related omics studies. Wei et al. utilized deep learning, specifically a simplified one-dimensional convolutional neural network (1DCNN), to analyze metallomic data from synchrotron radiation X-ray fluorescence (SRXRF) and build classification models of malignant pulmonary nodules without needing to quantify metal element concentrations (DOI: 10.1021/envhealth.4c00124). This method offers a less invasive, easier-to-handle alternative to traditional histopathological examination for lung cancer diagnosis and can serve as a “fingerprint metallome profile” to identify malignant pulmonary nodules.
Green chemistry can also benefit from AI/ML. Li et al. reported a framework, termed GPstack-RNN, to screen ionic liquids with high antibacterial ability and low cytotoxicity. Such a deep learning model can help accelerate the discovery of useful, safe, and sustainable materials (DOI: 10.1021/envhealth.4c00088). Meanwhile, Hao et al. discussed the potential of machine learning, particularly deep learning, to discover optimal green plastic additives (DOI: 10.1021/envhealth.5c00036).
The surge of environmental health related data brings research opportunities, as well as critical ethical concerns. Yu et al. proposed a checklist of ethical guidelines in data collection, data analysis, and data sharing in the AI era (DOI: 10.1021/envhealth.4c00273). They proposed some checkpoints such as clear labeling of simulated or augmented data, proper documentation of model architecture and hyperparameter optimization to track bias, and implementation of XAI techniques to improve interpretability. It is always important to promote responsible data practices that maximize the benefits of AI and big data while maintaining scientific integrity and protecting individual privacy.
In summary, this Special Issue covers various AI/ML-related topics including toxicity prediction, risk assessment, exposure estimation, environmental health, environmentally friendly material design, and data ethics. Ensemble model showed an impressive performance compared to single model and deep learning often achieved a better performance. XAI and other interpretable algorithms were introduced to investigate the mechanism of certain exposure processes. New strategies, tools, or frameworks based on AI/ML have been proposed to solve environmental health issues. We hope this Special Issue will inspire research insights and benefit more studies.
Views expressed in this editorial are those of the authors and not necessarily the views of the ACS.