Image created by Dr. Michael J. Miller
Have you ever wondered how researchers use artificial intelligence (AI) in environmental science? Certain AI computer models can learn from scientific data and operate without manual programming, in a process called machine learning. Machine learning models can identify patterns in environmental datasets and help scientists assess public health risks linked to environmental conditions.
One well-known public health risk is the rise in antibiotic-resistant bacteria. These bacteria carry genes called antibiotic resistance genes that protect them from antibiotic medicines. As the number of these genes increases, the medicines become less effective, and diseases spread faster.
Agricultural soils are the largest reservoirs of antibiotic resistance genes, because manure from livestock like cows and pigs often contains leftover antibiotics fed to them. When used as fertilizer, this manure transfers antibiotics into the agricultural soil, where bacteria develop antibiotic resistance genes.
To better understand how antibiotic-resistant bacteria will affect global health in the future, scientists want to develop faster ways to detect antibiotic resistance genes. However, current methods of detection are slow and fail to predict future trends. To address this problem, a group of scientists from Algeria used machine learning models to predict the prevalence of antibiotic resistance genes in soil microbes under future climate scenarios, and to identify the key environmental drivers and high-risk areas.
The team compiled existing microbial data from 3 public databases: the National Center for Biotechnology Information’s Sequence Read Archive, the Metagenomic Rapid Annotations using Subsystems Technology database, and the Joint Genome Institute’s Integrated Microbial Genomes & Microbiomes database. Their final dataset contained about 2,000 agricultural soil samples from 67 countries across 6 continents. These samples represented a wide range of soil characteristics, climate conditions, and different types of antibiotic resistance genes.
They integrated the microbial dataset with climate data, including global temperature and precipitation measurements from WorldClim, and land use data, including crop types, irrigation systems, and livestock density from the European Space Agency Climate Change Initiative Land Cover maps. The researchers also included data on future climate projections for 2050 and 2070 from the World Climate Research Programme based on low, medium, and high global greenhouse gas emission scenarios.
After preparing the datasets, the team set up 6 machine learning models that differ in terms of their data size, model complexity, analytical speed, and customization. These models included Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Support Vector Machine (SVM), Deep Neural Network (DNN), and Logistic Regression (LR).
They ran each model using different combinations of variables from their datasets, including soil properties, microbial community metrics, climate variables, and land use characteristics. To validate the models, the team separated the full dataset into 10 random subsets. They trained the models on 9 of the subsets, then recorded the result of the 10th subset. Each model repeated this process 10 times in a different order so that each subset was evaluated once in the results, using a technique known as stratified 10-fold cross-validation. The researchers scored each model on 5 performance metrics, including precision, sensitivity, and predictive power. Based on these results, they determined that LightGBM was the most accurate model.
They found that the LightGBM model identified soil temperature as the strongest environmental predictor of antibiotic resistance genes. At soil temperatures above 18°C (about 64°F), the number of antibiotic resistance genes increased dramatically. The model also identified soil acidity, organic carbon content, moisture, and annual precipitation as influential variables, accounting for 71% of the increase in antibiotic resistance genes under future climate scenarios. Likewise, the LightGBM model predicted that the number of high-risk areas would increase by 35% under the high-emissions scenario, with South Asia, Sub-Saharan Africa, and Mediterranean Europe as the most vulnerable regions.
Based on the LightGBM model output, the team concluded that soil conditions will change to favor antibiotic resistance genes under all future climate scenarios. They also suggested that LightGBM is a fast and effective model for environmental predictions that could be used to assess other climate change risks such as natural disasters. They recommended that future researchers explore how to apply machine learning methods to early warning systems and utilize machine learning results in environmental management decisions.
Reference
Meriem Kenzi, Meriem Benbernou, Hadjer Khelifa, Hadja Fatima Tbahriti, Machine learning-based prediction of antibiotic resistance gene distribution in agricultural soils under different climate change scenarios, Science of The Total Environment, Volume 1042, 2026, 181905, ISSN 0048-9697, https://doi.org/10.1016/j.scitotenv.2026.181905.
Abstract
Antibiotic resistance genes (ARGs) in agricultural soils represent a major public health concern, as climate change is believed to augment their dissemination and abundance. Understanding the impact of future climate change scenarios on ARG abundance is essential to implement predictive and proactive One Health strategies. In this study, a total of 2301 soil samples from 67 countries across six continents were compiled from three global metagenome databases, namely NCBI SRA, MG-RAST, and JGI IMG/M. Six machine learning models, namely LightGBM, XGBoost, Random Forest, Support Vector Machines, Deep Neural Networks, and Logistic Regression, were used to predict ARG distribution patterns in agricultural soils, and their performance was evaluated using stratified 10-fold cross-validation with metrics such as AUC-ROC, precision, recall, F1 score, and Matthews Correlation Coefficient. WorldClim 2.1 and CMIP6 models were used to project ARG distribution under three Representative Concentration Pathway scenarios, namely RCP 2.6, RCP 4.5, and RCP 8.5, for the years 2050 and 2070. The LightGBM model achieved the best predictive performance, with an AUC-ROC of 0.957 (95% CI: 0.951–0.963), substantially higher than that of the other models, while the Deep Neural Networks model achieved an AUC-ROC of 0.891. The LightGBM model demonstrated high stability across cross-validation folds, with minimal fold-to-fold variance, defined as the standard deviation of AUC-ROC scores across the 10 folds (SD = 0.008). SHAP feature importance analysis identified soil temperature, pH, and organic carbon content as the top three factors influencing ARG relative abundance, with SHAP values of 0.342, 0.287, and 0.251, respectively. Annual precipitation and soil moisture level were also identified as significant contributors to ARG distribution. SHAP dependency plots revealed critical thresholds for ARG relative abundance, with a sharp increase observed independently when soil temperature exceeds 18 °C and when soil pH drops below 6.5. Furthermore, a non-linear accelerating increase in ARG abundance risk was observed as climate change intensity worsened across scenarios. Projections for future climate change scenarios indicate a potential 34.7% increase in high-risk ARG zones by the year 2070, with the largest changes expected in South Asia, Sub-Saharan Africa, and Mediterranean regions. Paired t-tests revealed significant differences in performance among all models (p < 0.001). These findings demonstrate that gradient-boosting methods such as LightGBM outperform deep learning approaches for ARG prediction from soil microbiome data, offering higher accuracy and interpretability. As climate change is projected to increase ARG risks in a non-linear manner, the development of climate-adaptive agricultural practices and global surveillance systems is urgent. This framework provides actionable risk-mapping tools to support precision farming and region-specific policy interventions within the One Health approach.