Firing rate control

These are descriptions of ideas related to simulation average firing rate control and possible future directions of such work.

The article’s approach

In the article, the archetype simulated cells projects were based on recreating real cells from different animals and/or different grid module scales. The purpose of this was to avoid the scope of work attempting to have precise control over average firing rates of grid cells recorded in the same animal, in the same naural layer (and grid cell module by extension), and recorded at the same time. This level of firing rate control was considered out-of-scope of the current work because it may involve biophysics and/or cellular mechanism beyond that which is included in the current model. Additional details explaining this information will be covered here.

The average firing rate statistical analyses in the article are not affected by trying to precisely control the firing rate in the above described same layer case. The reason for this is that the archetype projects aimed to recreate cells from different animals and/or grid modules. The simulation included elements adapted from existing prior models such as uniform synaptic conductance constant levels from head direction and speed input signaling into grid cells. This can be considered to be a conductance constant value set on a neuron-type basis as opposed to an individual neuron basis. This uniform conductance conceptually creates somewhat similar average firing among grid cells within the same layer. For instance, if these synaptic conductance constant values varied further from cell-to-cell, but was configured in a way to sustain grid cell pattern firing, then more variation in average firing rates between cells could occur.

Noticable average firing rate differences have been observed to occur in simulations of grid cells of the same animal, the same neural layer, and same recording time. This is similar to an extent to such grid cell conditions observed in real cell data. For instance, real grid cells in the same animal (id: ArchTChAT_22), recorded at presumably the same time, with the property of having intermediate grid scales have been observed to have an average firing rate range of 0.23 Hz to 2.4 Hz (N=9 cells). A project that was based on recreating one of those cells in particular generated grid cells with a firing rate range of 1.1 Hz to 2.9 Hz (N=1600 cells). The average firing rate ranges are somewhat similar even with the simulation work having properties such as the uniform conductance constant levels. The various properties included in the simulation project cause the average firing rate variation to occur. This range similarity indicates a potential of the simulation being able to be refined to better match the average firing rates of the real cells in a closer way.

Different simulated experiments have been observed to have different average firing rate ranges. For example, one experiment had a range of ~0.9 Hz (N=~1600) while another had a range of ~11.0 Hz (N=1600). In the simulation projects, in general, projects that include higher average firing rates typically have greater ranges then those with only low average firing rates. In general, statistically, the ranges can be somewhat similar between simulated and real cells. However, simulation challenges exist in trying to precisely control average firing rates of more than one specific simulated grid cell in the same layer that is recorded at the same time. This is why the article did not attempt to attain this level of control at this stage of simulation work. Precise control over grid cell average firing rates in different animals and/or modules was achived as described in the article. Something that helps that control is the ability to tailor conductance constant levels, and other model parameters, to values that facilitate the creation of targeted firing rate levels. Some of those parameters are uniform for the whole neural layer (e.g., set by neuron type) in the current simulation but in future work can have individual cellular-level values.

Future potential directions

Multiple methods may be used to achive more specific control over the average firing rates of cells within the same layer that are recorded at the same time. Some methods that could be included to enhance such control include integrating neuromodulation into the model. Other methods include adding to the model more complex connectivity or additional neuron types. More individual cell specific synaptic conductance constant values is a form of connectivity that could be added. Long term plasticity could be included in the model to aid in the goal. Possibly contections with weaker synaptic strengths relative to other grid cells could connect to the grid cell that has a target lower firing rate. When the attractor bump moves over that targeted neuron the firing levels could be reduced relative to other neurons because of less signal being sent to it. A challenge may exist with balancing such a distribution of weaker and stronger connection in a way that accomidates the standard operations of attractor bumps moving throughout a continuous attractor network of grid cells. Machine learning optimization of simulation parameter exploration toward achiving specific firing rates could aid in finding settings that produce desired results. These approaches can also benefit from greater evidence reporting from animal studies. For example, targeted explorations of neuromodulation levels in grid cells can produce data that refines what activity is targeted by the simulation. The potential future approaches listed here are a few of the many way that improved firing control could be accomplished.