Rare and endangered animals are often among the species most vulnerable to chemical pollution, yet scientists face a difficult problem when trying to assess those risks. Conventional toxicity experiments may require large numbers of organisms, making them impractical and ethically inappropriate for species with limited populations.
A new study published in New Contaminants demonstrates how machine learning can help predict the effects of pollutants on endangered species while reducing the need for direct biological testing. The researchers developed a machine learning enhanced quantitative structure activity relationship model, known as ML-QSAR, using the rare gudgeon (Gobiocypris rarus) as a case study.
The rare gudgeon is a small freshwater fish native to China's Yangtze River Basin. Because of its restricted distribution and sensitivity to environmental changes, it has been designated a rare and endangered species.
"Our goal was to develop a practical method for estimating chemical toxicity in species for which experimental data are limited and conventional testing is difficult," said corresponding author Ying Wang. "By combining molecular information with the life stage of the fish, the model can provide evidence to support conservation and pollution management."
The researchers compiled available acute and chronic toxicity data for the rare gudgeon and calculated more than 1,800 molecular descriptors representing chemical properties such as structure, electronic behavior, polarity and potential interactions with biological targets. They also included the developmental stage of the fish, distinguishing among embryos, juveniles and adults.
Six machine learning algorithms were evaluated, including random forest, support vector machine, neural network and generalized linear models. The random forest model produced the strongest overall performance, achieving a coefficient of determination of 0.99 for acute toxicity and 0.93 for chronic toxicity.
The analysis also revealed an important difference between short-term and long-term chemical effects.
Life stage was one of the most influential factors in predicting acute toxicity. Embryonic and juvenile fish were generally more sensitive to many pollutants because their metabolic and detoxification systems are still developing. However, the researchers found that this pattern could vary among chemical groups. Adult fish, for example, may retain some PFAS compounds for longer periods because these substances bind strongly to proteins.
For chronic toxicity, molecular interaction descriptors were more important than life stage. These descriptors reflected properties such as ionization potential, polarizability and the spatial arrangement of atoms, all of which can influence how chemicals move through water, accumulate in organisms and interact with biological molecules.
The researchers applied the best-performing models to 73 pollutants reported in the rare gudgeon's habitat, including per- and polyfluoroalkyl substances, commonly called PFAS. Environmental concentration data were available for 12 PFAS compounds.
The calculated risk quotients for those PFAS compounds were well below 1, suggesting that currently reported concentrations pose a low immediate ecological risk to the rare gudgeon. Nevertheless, the authors emphasized that this result should not be interpreted as evidence that PFAS pollution is harmless.
PFAS compounds are highly persistent and can accumulate through food webs. Their concentrations may also change with industrial activity, seasonal conditions and the growing use of replacement chemicals. The researchers therefore recommend long-term monitoring of PFAS distribution and bioaccumulation in the fish's habitat.
The study provides a non-testing framework that could be adapted for other threatened aquatic species. Future research will need to expand available toxicity datasets, examine chemical mixtures and improve predictions for metals and newly emerging contaminants.
By linking chemical structure, developmental biology and machine learning, the approach could help conservation managers identify potential pollutant threats before endangered populations experience irreversible harm.
===
Journal reference: Wang Y, Wang X, Zhou Y, Cheng Y, Li X, et al. 2026. Toxicity prediction and ecological risk assessment of new contaminants to rare and endangered species using machine learning-QSAR: a case study of conserving Gobiocypris rarus in the Yangtze River Basin. New Contaminants 2: e015 doi: 10.48130/newcontam-0026-0010
https://www.maxapress.com/article/doi/10.48130/newcontam-0026-0010
===
About the Journal:
New Contaminants (e-ISSN 3069-7603) is an open-access journal focusing on research related to emerging pollutants and their remediation.
