Cross-species behavior analysis with attention-based domain-adversarial deep neural networks

Area of Science:

• Neuroscience
• Computational Biology
• Animal Behavior

Background:

• Studying human diseases in model organisms is crucial for understanding underlying mechanisms.
• Analyzing animal locomotion data is challenging due to variations in spatial and temporal scales across species.
• Conventional statistical methods are insufficient for extracting cross-species knowledge from locomotion data.

Purpose of the Study:

• To propose a procedure for automatically discovering shared locomotion features across different animal species.
• To develop a method that overcomes the limitations of traditional statistical analyses in cross-species locomotion studies.
• To create a human-interpretable framework for understanding complex animal movement patterns.

Main Methods:

• Utilizing domain-adversarial deep neural networks to identify shared locomotion features.
• Incorporating an attention mechanism to explain hidden cross-species features within the neural network.
• Formulating human-interpretable rules from the neural network's findings.
• Validating discovered features using statistical tests.

Main Results:

• Successfully identified locomotion features shared among humans, mice, and worms with dopamine deficiency.
• Demonstrated the versatility of the proposed procedure across evolutionarily distant species.
• The attention mechanism provided interpretable insights into cross-species locomotion patterns.

Conclusions:

• The developed deep learning procedure effectively identifies conserved locomotion features across species.
• This method offers a powerful tool for comparative analysis of animal behavior and disease mechanisms.
• The approach facilitates the translation of findings from model organisms to human health.