Intelligent larval zebrafish phenotype recognition via attention mechanism for high-throughput screening

Area of Science:

• Developmental biology
• Ecotoxicology
• Computational biology

Background:

• Larval zebrafish phenotypes are key indicators in ecotoxicology and safety assessments.
• Identifying these phenotypes is labor-intensive and requires expertise.
• Automated methods are needed to streamline phenotype analysis.

Purpose of the Study:

• To develop and validate a deep learning model for automated larval zebrafish phenotype classification.
• To improve the efficiency and accuracy of identifying phenotypic defects in zebrafish.

Main Methods:

• Proposed RECNet, a deep network model integrating attention mechanisms and residual structures.
• Utilized mixup data augmentation and a dataset of 6805 larval zebrafish phenotype images.
• Applied the model to four-class and seven-class (mixed abnormalities) classification tasks.

Main Results:

• Achieved high accuracy (0.949) and F1-score (0.966) in the four-class task.
• Attained 0.913 accuracy and 0.847 mean average precision in the seven-class anomaly task using DFBLoss.
• Demonstrated superior performance on a new test dataset and with a larger training dataset than previous studies.

Conclusions:

• The RECNet model offers a promising automated solution for larval zebrafish phenotype analysis.
• This approach can significantly support research in toxicology, ecotoxicology, and safety assessment.
• Facilitates more efficient and accurate data acquisition in zebrafish laboratories.