Inpainting the Neural Picture: Inferring Unrecorded Brain Area Dynamics from Multi-Animal Datasets

Area of Science:

• Systems neuroscience
• Computational neuroscience
• Neuroimaging analysis

Background:

• Understanding brain area interactions is key in systems neuroscience.
• Simultaneous recording of all brain areas is often infeasible within a single experiment.
• Multi-animal datasets offer potential for studying broader neural interactions.

Purpose of the Study:

• To develop a method for inferring the dynamics of unrecorded brain areas using multi-animal data.
• To enable multi-area interaction analyses despite limitations of individual recordings.
• To leverage masked autoencoding for reconstructing neural activity patterns.

Main Methods:

• Introduced NeuroPaint, a masked autoencoding model.
• Trained the model across multiple animals with overlapping recorded brain areas.
• Evaluated performance on synthetic and real-world Neuropixels datasets.

Main Results:

• NeuroPaint successfully reconstructs the dynamics of unrecorded brain areas.
• The model leverages shared structure across individuals with partial observations.
• Enables multi-area analyses beyond the scope of single experiments.

Conclusions:

• Masked autoencoding across animals is effective for inferring neural dynamics.
• NeuroPaint overcomes limitations of incomplete neural recordings.
• Facilitates more holistic understanding of brain-wide interactions.