Allowing user defined input function in evoked drives

[zhongyileon]

Hey there!
I wanted to ask or discuss if there is an option in HNN to allow a user defined input function when adding an evoked drive? As I understand there is currently only the option for a gaussian shaped drive (and poisson and bursty). We are trying to model mid latency responses of the auditory system using a up-chirp stimulus and the data could suggest an alpha shaped function for the input and we wanted to test this hypothesis.

The experimental data (blue) has a much steeper rise and a flatter top that is difficult to model with a gaussian input. Any help or info would be appreciated!

[asoplata]

Hello @zhongyileon !

You are correct that there is currently no way to insert a custom input function shape into our drives. However, we may still be able to suggest or help build a compatible drive for your research.

To do so, we first need to understand your intended stimulus. Can you please describe the input you are imagining in more detail?

A few thoughts about what you've described:

• You mention an "alpha-shaped function". First, just to be clear, all of our drives (except for the "tonic bias") are composed of many driving spikes, and those spikes are distributed according to a statistical distribution in time, such as a gaussian/poisson/etc. When you say an alpha-shaped function, if you mean something similar to the description given at this webpage https://www.sas.upenn.edu/LabManuals/BBB251/NIA/NEUROLAB/HELP/alphasyn.htm , then that is often used to refer to the shape of an individual synaptic event. The shape of a single synaptic event is not what is shown in our input histograms, which are shown in green and red in the figure you've provided. You probably understand this, but I just want to make it very clear and avoid possibility of confusion: our external drives are made up of very many spikes, not a single synaptic event. You could theoretically use the same alpha-function-like shape to define the distribution of the many spikes comprising your drive, but that would be independent of the shape of the individual synaptic conductance changes themselves. We do not have code to do that, currently. I believe you're talking about this case: where an alpha-function is used as the shape/distribution of many spikes in your drive. (I apologize if, by trying to reduce confusion, I have only increased it)

• You mention an "up-chirp" stimulus. I'm familiar with chirp stimuli such as here https://pmc.ncbi.nlm.nih.gov/articles/PMC3188598/figure/f3/ . Though I'm not specifically familiar with the term "up-chirp" specifically, I assume that just means a chirp where the frequency goes up over the course of the signal. However, in this case, we're not sure what you want your input to look like, since this could be implemented in a couple different ways. Some possibilities are listed below, but we do not currently have code for any of these:

  • Perhaps you want a drive that consists of many spikes, but where the frequency of the spike rate is subject to a "chirp" and increases rapidly. This is my best guess about what you mean.
  • Perhaps you want a stimulus that does not consist of spikes, but instead consists of a chirp-like direct injection of current into cells, such as in this figure: https://www.frontiersin.org/files/Articles/703407/fncel-15-703407-HTML-r1/image_m/fncel-15-703407-g002.jpg . I don't think that this is what you want, but am mentioning it for clarity.
  • Another possibility is that you do not actually want to model the "chirp" part of the stimulus in HNN at all, and instead you only want to model a "downstream representation" (my words) of the stimulus, which you believe (for scientific reasons) resembles an "alpha-shaped function".

Depending on what you want, I have confidence that we will be able to help craft the stimulus you want.

Additionally, I will mention that we currently have a new feature in development that may be very helpful for your case: replaying custom spike-train data as its own drive. This would allow you to hand-craft your own exact, custom spike train (or matrix of multiple spike trains) and then replay them as drives into your network. This feature is fully implemented, but currently it is only available on our development branch: the good news is it will be included in the next version release, which will be published before the end of October (within the next 30 days). A tutorial for this in-development feature is available here: https://jonescompneurolab.github.io/hnn-core/dev/auto_examples/howto/plot_replaying_spike_data_as_input.html#sphx-glr-auto-examples-howto-plot-replaying-spike-data-as-input-py . If you are interested in this, then you are welcome to install and use it by installing the latest development version of HNN, which we provide instructions for in the pip Source Installation section of our Installation Guide here https://jonescompneurolab.github.io/textbook/content/01_getting_started/installation.html . If you decide to go this route and run into any installation issues, please let me know and I'll be happy to assist you. Installing a development version of HNN is slightlymore difficult than a regular installation, but we have worked hard to make our instructions as easy to understand as possible.

Best,
Austin