Effective computations for hippocampal place cell phenomena in sparse untrained random networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Mammalian brains process spatial information using complex network mechanisms.
• Existing models struggle to explain hippocampal place cell phenomena like remapping and representational drift.
• The necessity of specific network connectivity and synaptic plasticity for spatial processing remains debated.

Purpose of the Study:

• To investigate whether specific connectomes and synaptic plasticity are essential for diverse place cell phenomena.
• To examine the explanatory power of a randomly tuned network with feedback inhibition for spatial information processing.
• To determine if simpler, biologically plausible network architectures can account for complex hippocampal activity.

Main Methods:

• Utilized a randomly tuned network model with feedback inhibition (DivSparse).
• Simulated network activity and analyzed positional tuning, place fields, selectivity, and remapping.
• Incorporated synaptic plasticity at network connections (not inputs) to model additional phenomena.

Main Results:

• A random network with non-plastic connections successfully explained positional tuning, single/multiple place fields, mixed selectivity, and remapping.
• Sparse, normalized excitatory activity was crucial for explaining these phenomena.
• Enabling plasticity only at network connections accounted for overdispersion, representational drift, and memory tagging.

Conclusions:

• A simple, randomly connected network with sparsifying inhibition (DivSparse) can explain numerous place cell phenomena.
• Specific connectomes and input-based learning may not be necessary for robust spatial representations.
• Biologically plausible, simple architectures can support flexible and spontaneous representations of spatial experience.