EmbryoNet: using deep learning to link embryonic phenotypes to signaling pathways

Area of Science:

• Developmental Biology
• Machine Learning
• Bioinformatics

Background:

• Signaling pathways are crucial for embryonic development, but classifying resulting defects is complex and lacks standardization.
• Expert knowledge is currently required for accurate phenotyping, hindering large-scale analysis.

Purpose of the Study:

• To develop an automated phenotyping system using machine learning to classify developmental defects in zebrafish.
• To identify and classify defects caused by the loss of function in seven major vertebrate signaling pathways.

Main Methods:

• Trained a deep convolutional neural network, named EmbryoNet, on over 2 million zebrafish images.
• Integrated EmbryoNet with a model of time-dependent developmental trajectories for precise defect classification.
• Applied automated phenotyping in high-throughput drug screening.

Main Results:

• EmbryoNet accurately identifies and classifies phenotypic defects in zebrafish signaling mutants.
• The system demonstrates high precision in classifying defects related to seven major signaling pathways.
• Successfully resolved the mechanism of action for pharmaceutical substances in drug screens.

Conclusions:

• Automated phenotyping with EmbryoNet offers a standardized and unbiased approach to studying developmental signaling pathways.
• This AI tool has broad applications in developmental biology, cross-species analysis, and pharmaceutical research.
• The freely available image dataset facilitates further research in automated embryo phenotyping.