Dev-ResNet: automated developmental event detection using deep learning.

Area of Science:

• Developmental Biology
• Machine Learning
• Bioimaging

Background:

• Identifying developmental events is crucial for understanding biological processes across species and environments.
• Current methods for detecting developmental events are challenging, limiting experimental scale and reproducibility.
• Bioimaging is widely used in developmental biology, but analyzing the data remains a bottleneck.

Purpose of the Study:

• To introduce Dev-ResNet, an efficient 3D convolutional neural network for detecting developmental events.
• To demonstrate Dev-ResNet's capability in identifying events with spatial and temporal features.
• To provide a scalable and reproducible method for analyzing early life stage development.

Main Methods:

• Development of Dev-ResNet, a small and efficient 3D convolutional neural network.
• Application of Dev-ResNet to detect 10 diverse functional events during embryonic development in Lymnaea stagnalis.
• Validation of Dev-ResNet's accuracy in identifying event onsets and thermally induced timing changes.

Main Results:

• Dev-ResNet successfully detected the onset of all 10 tested developmental events.
• The tool accurately identified thermally induced decoupling of developmental event timings.
• Dev-ResNet demonstrated high efficacy in analyzing spatial and temporal features of developmental events.

Conclusions:

• Dev-ResNet offers a powerful and efficient solution for delineating developmental events in early life stages.
• The deep learning approach enhances the scale, reproducibility, and throughput of experimental biology.
• Dev-ResNet has broad applicability in bioimaging and developmental biology research, with provided resources for implementation.