Images of daphnids (control and exposed to NMs) over multiple generations, scored by experts as toxic or non-toxic and the resulting deepDaph predictions

Background

This study showcases a pioneering application of deep learning methodologies in ecotoxicology, aimed at facilitating hazard assessment and safer design of engineered nanomaterials (ENMs). The research hinges on a high-quality dataset comprising microscopic images of Daphnia magna exposed to various ENMs, collected systematically under controlled conditions.

The Dataset: A Cornerstone of Nanoinformatics

Our dataset, which will be openly accessible on Zenodo, serves as a foundational resource for the ecotoxicology community. It contains high-resolution images tagged with intricate details like malformations, tail lengths, lipid concentrations, and lipid deposit shapes. Researchers can use this exhaustive dataset to train a variety of predictive models for diverse applications.

Methodology

We employ two different deep learning architectures to process the dataset. These architectures automatically detect malformations and assess the impact of ENMs on D. magna by classifying various biological structures based on lipid densities.

Results and Validation

The developed models demonstrate high statistical validation, confirming their prediction accuracy on external D. magna images. Our dataset and the associated models not only accelerate manual procedures but also pave the way for automated, high-throughput analyses in ecotoxicology.

Future Prospects

The dataset holds the potential to extend investigations into predicting the impacts on future generations from parental exposures, thus reducing the time and cost of multi-generational toxicity assays.