Reusable modular architecture enables flexible cognitive operations.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Complex behaviors are theorized to emerge from the integration of simpler cognitive elements.
• Artificial neural networks exhibit task flexibility through modular neuron combinations, but empirical brain evidence is scarce.

Purpose of the Study:

• To investigate the empirical evidence for reusable modular neural networks in the brain.
• To explore how the brain flexibly combines neural components for complex cognitive tasks.

Main Methods:

• Mice performed a delayed match-to-sample with delayed report (DMS-dr) task.
• Analysis of neural activity subspaces in medial prefrontal cortex (mPFC) and posterior parietal cortex (PPC).
• Clustering analyses to identify functionally distinct neuron clusters.
• Recurrent neural network modeling to validate computational roles and modularity.

Main Results:

• Identified reusable neural activity subspaces specialized for stimulus processing and memory maintenance.
• Demonstrated reuse of these subspaces for new stimuli and different memory types.
• Clustering revealed functionally distinct neuron groups in mPFC and PPC supporting these subspaces.
• Neural network models confirmed that silencing specific neuron clusters impaired specific computations.

Conclusions:

• The brain utilizes a modular and reusable organization of neural networks for complex cognition.
• Findings support the hypothesis that the brain flexibly reuses computational components.
• Provides empirical evidence bridging computational predictions and neural mechanisms.