Connectome-constrained networks predict neural activity across the fly visual system

Area of Science:

• Computational neuroscience
• Systems neuroscience
• Connectomics

Background:

• Neural circuit connectivity is measurable, but neuron dynamics are not.
• Understanding neural computation solely from connectivity is challenging.
• Sparse connectivity is a universal feature of biological neural networks.

Purpose of the Study:

• To predict neural activity and understand neural computation using only connectivity data.
• To develop a strategy for generating hypotheses about neural circuit mechanisms.
• To investigate the role of sparse connectivity in successful prediction.

Main Methods:

• Constructed a model neural network using experimentally determined fruit fly optic lobe connectivity.
• Optimized unknown single-neuron and single-synapse parameters using deep learning techniques.
• Validated model predictions against experimental neural activity measurements.

Main Results:

• The model successfully predicted neural activity for visual motion detection.
• Model predictions aligned with experimental data across 26 studies.
• The prediction strategy proved more effective with sparse neural connectivity.

Conclusions:

• Connectivity data, when combined with deep learning, can predict neural activity and inform neural computation understanding.
• This study provides a powerful strategy for generating testable hypotheses about neural circuit function.
• Sparse connectivity enhances the ability to predict neural circuit behavior from connectomic data.