Related Experiment Video
Updated: Jan 15, 2026
02:52
Small Molecule Screening and Toxicity Testing in Early-stage Zebrafish Larvae
Published on: March 7, 2025
2.0K
Deep Learning-Enabled Unbiased Precision Toxicity Assessment of Zebrafish Organ Development
Mengyu Wang1,2, Wen-Xiong Wang1,2
1School of Energy and Environment, State Key Laboratory of Marine Environmental Health, City University of Hong Kong, Kowloon, Hong Kong, China.
Environmental Science & Technology
|October 15, 2025
Summary
A new deep learning model precisely assesses toxic effects in biological images, overcoming limitations of traditional methods. This AI approach reveals previously undetectable size-dependent toxicity in zebrafish, improving accuracy and efficiency in toxicological analysis.
Area of Science:
- Toxicology
- Biomedical Science
- Environmental Health
- Artificial Intelligence
Background:
- Traditional toxicology methods struggle with sensitivity and bias due to reliance on macroscopic endpoints and manual image analysis.
- Precise assessment of toxicological effects is crucial for biomedical and environmental health evaluations but remains a significant challenge.
Purpose of the Study:
- To develop an automated deep learning approach for precise and objective toxicological analysis.
- To establish a general framework for unbiased assessment of toxic effects using artificial intelligence.
- To evaluate the model's ability to detect subtle, size-dependent toxicity in developing organisms.
Main Methods:
- Developed a U-Net based deep learning model for automated, pixel-level image segmentation and morphological quantification.
- Applied the model to analyze thousands of biological images, achieving high speed (1 min per batch) and eliminating subjective bias.
- Utilized the model to assess size-dependent developmental toxicity induced by silver ions (Ag+) and silver nanoparticles (AgNPs) in zebrafish.
Main Results:
- The U-Net model demonstrated efficient and unbiased analysis of biological images, performing pixel-level segmentation and morphological quantification.
- Successfully distinguished size-dependent developmental toxicity of Ag+ and AgNPs (15 nm, 100 nm) in zebrafish.
- Identified previously undetectable size-dependent and organ-specific toxicity disparities in the photoreceptor cell layer, inner plexiform layer, skeletal muscle, and spinal cord.
Conclusions:
- The automated deep learning approach offers a significant improvement in accuracy, efficiency, and reproducibility for toxicological assessments.
- The framework provides a scalable application for precise toxicological evaluations, including standardized imaging analysis.
- This method holds potential for assessing the toxicity of emerging materials and contaminants, advancing environmental and biomedical health assessments.
