The digital sphinx: Can a worm brain control a fly body?

Area of Science:

• Neuroscience
• Computational Biology
• Biophysics

Background:

• Animal intelligence arises from integrated neural and body dynamics, not just brain computation.
• Virtual animals created with deep reinforcement learning (DRL) offer a way to study embodied intelligence.
• Gaps in biological parameters necessitate data-driven approaches like DRL for modeling.

Purpose of the Study:

• To investigate the potential pitfalls of interpreting virtual animal models.
• To highlight the importance of biological grounding in connectome-body simulations.
• To caution against overinterpreting behavioral mimicry as biological fidelity.

Main Methods:

• Constructed a virtual chimera using a nematode worm (C. elegans) connectome and a fruit fly (Drosophila) biomechanical model.
• Integrated sensory input from the fly body to the worm connectome.
• Trained an artificial neural network using DRL to map worm neural outputs to fly motor controls.

Main Results:

• The resulting digital chimera exhibited highly realistic fly walking behavior.
• Despite behavioral realism, the model lacked biological fidelity and offered no insights into either organism.
• Demonstrated that behavioral mimicry can be achieved independently of biological accuracy.

Conclusions:

• Connectome-body models can achieve behavioral fidelity without biological fidelity, posing a risk of overinterpretation.
• Virtual animals are valuable research tools only when their components and interfaces are biologically grounded.
• Emphasizes the need for careful validation and biological relevance in computational models of animal behavior.