Deep learning image recognition enables efficient genome editing in zebrafish by automated injections

Area of Science:

• Zebrafish model organism research
• Developmental biology
• Biomedical engineering

Background:

• Microinjection is a key technique in zebrafish research for genetic manipulation and substance delivery.
• Current methods can be time-consuming and require significant expertise.

Purpose of the Study:

• To develop and validate a machine learning-based software for automated, high-speed microinjection in zebrafish embryos.
• To enhance the efficiency and throughput of genetic manipulation techniques in zebrafish.

Main Methods:

• Development of a deep learning software (Inception v3) for embryo identification and precise injection site localization.
• Utilized Graphics Processing Unit (GPU) for rapid processing (<100 milliseconds per injection cycle).
• Tested automated injection efficiency using morpholino, CRISPR/Cas9, and DNA constructs.

Main Results:

• The software achieved 93% accuracy in distinguishing embryos and predicted injection sites within 42 μm of manual annotations.
• Automated injection efficiency for morpholinos was comparable to manual methods (~80%).
• CRISPR/Cas9 and DNA construct injection efficiencies matched those of experienced researchers, with higher overall yield.

Conclusions:

• The developed machine learning software significantly increases the speed and efficiency of zebrafish microinjection.
• This automated system facilitates high-throughput genetic screening and gene editing (CRISPR/Cas9) for biomedical research.
• Enables rapid gene knockout and knock-in studies to investigate gene function and biological pathways.