Modelling other mechanisms that affect Current/Voltage

[kgeorgiadi]

Hello,

I am new to NEURON, I have recently started a PhD and we are trying to model certain things besides the typical ion concentrations that affect the firing rate of neurons and other properties of a neural network. My background is in computer science and I've learnt a few things in neuroscience but very basic.

Some of the things I would like to model are pathogenic proteins(i.e. dementia diseases), nutrients that may affect the neuron, etc. I have looked up to half way into these tutorials: http://www.anc.ed.ac.uk/school/neuron/t ... /tutD.html

It seems like I may be able to "hack" model these things as "ion" concentrations in the intra and extracellular parts of the membrane. Since NEURON is so old and vast and the code is too much to read all of it and understand everything, I was wondering if you could help guide me a bit as to what I should read to do something like this. Basically I think I primarily want 3 things:

Somehow model that outside of neuron #1 there is say x amount of pathogenic protein(e.g. beta-amyloid) and inside of it y amount. This y amount can affect the current/ion/voltage buildup and affect the firing rate through some equation that I will define(how?). The pathogenic proteins can also propagate through synapses from neuron #1 to another neuron(can this be modelled during the simulation somehow easily?). Lastly, is it possible to change things during the simulation? For example, if there is a high amount of the pathogenic protein inside the neuron, to somehow model that this neuron "dies" and its synapses do not work anymore?(i.e., the simulation will check at every timestep whether the concentration of this protein inside each and every neuron is above a threshold and then do something)

Finally and perhaps most importantly, can I do all of these as a "hack" through NMODL and hoc code or will I need to change the source code of NEURON for some of these?

Any help is greatly appreciated! Thank you very much,
Konstantinos

[ted]

NEURON is capable of a lot more than you imagine. Of course, so are C and C++, and the latter have the advantage of offering complete freedom to implement whatever you like without having to learn anything about an existing code base. However, before diving into the weeds of implementation, it is important to have a clear conceptual model. To paraphrase an old saying, there are many ways to skin a cat, but in any particular situation only a few will be graceful and efficient.

The conceptual model exists not in a computer, but in your mind, and it should include those features of a neuron or network of neurons that are relevant to your hypothesis of pathogenesis. It will dictate what goes into the computational implementation.

Once you have created a computational implementation of your conceptual model, the next question is what do you do with it? In other words, what will be your computational experiment?

Some of the things I would like to model are pathogenic proteins(i.e. dementia diseases), nutrients that may affect the neuron, etc.

You have now left NEURON's domain of utility and moved into the realm of modeling molecular structure or molecular pathways. Or did you mean you want to study how such things affect the operation of neurons, i.e. their effects on the generation, propagation, and interaction of electrical and chemical signals?

Let's suppose you mean the latter. Those agents generally have their effects over the course of weeks or months, if not years. You probably don't intend to run simulations that span months or years of model time. It might make sense to start with a "normal" in which there are one or more parameters that are susceptible to some factor X. Presumably X affects the values of certain parameters, and you know what the parameter values should be normally and what they will be after treatment with X. The computational experiment would consist of running two simulations: one in which the model's parameters have their normal values, and another in which their values have been perturbed.

Or does your hypothesis state that X has an acute effect (on the scale of milliseconds or seconds, maybe even minutes or hours)? If so, then it makes sense to run a simulation that starts out with normal parameters, then at some time the parameters are perturbed.

[kgeorgiadi]

Thank you for the reply! I am a bit confused with what you are saying here:

Let's suppose you mean the latter. Those agents generally have their effects over the course of weeks or months, if not years. You probably don't intend to run simulations that span months or years of model time. It might make sense to start with a "normal" in which there are one or more parameters that are susceptible to some factor X. Presumably X affects the values of certain parameters, and you know what the parameter values should be normally and what they will be after treatment with X. The computational experiment would consist of running two simulations: one in which the model's parameters have their normal values, and another in which their values have been perturbed.

Or does your hypothesis state that X has an acute effect (on the scale of milliseconds or seconds, maybe even minutes or hours)? If so, then it makes sense to run a simulation that starts out with normal parameters, then at some time the parameters are perturbed.

The basic idea is to currently try to work on a small network (10-200 neurons) rather than an entire brain and hopefully later during the PhD we can do something like this in a petri dish too to get validation and refine our modelling. Therefore I assume that the effects will not take as long but could possibly be in the order of minutes or hours for each neuron.

Right now I would have tried something like this:

Run one experiment for say 30 minutes where there are no pathogenic proteins.
Then run an experiment with the exact same network structure but add pathogenic proteins.

In both cases I would try to save the Voltage time curves for all neurons and then probably use some other software to detect when there was an action potential. Then I could also see which neurons end up stop firing or fire less often due to the effects of the pathogens.

To be honest right now we are not even sure how exactly we will model everything or what we will be looking for. After all, even in neuroscience there are more hypotheses about how these things work than clarity. I think that is the reason my supervisor wanted to try a computational modelling approach, so as to try various experiments and see which things seem to cause similar effects as in the real world.(i.e. the petri dish validation we may get to do later).

I looked over the rest of the tutorial that I linked before. I am wondering if through NMODL you are basically defining your own variables and section(which you can then "insert" to another section. Then I can define functions on how these variables change over time, so it would seem there are easy ways for me to do most things I would want to implement. Is this correct?

I am still curious though as to what you think I should read on if I wanted to try to do all of the things I listed before with just the hoc and NMODL languages?

[ted]

Thank you for the reply! I am a bit confused with what you are saying here

I don't know how I could have been clearer.

A few key points to consider before proceeding further:
1. A computational model of a cell is only an implementation of a conceptual model of a cell in a form that is suitable for simulation by a computer.
2. A conceptual model of a cell includes only those properties of a real cell that the model implementer decides to include. It is a simplified abstraction of a real cell.
3. Selection of the properties that are included in a given conceptual model is guided by the hypothesis that drives the research. If there is no hypothesis, there is no rational basis for deciding which properties to include in the conceptual model, and no rational basis for implementing a computational model.
4. Design of the computational experiments that are used to study the model is also guided by the hypothesis that drives the research. If there is no hypothesis, there is no basis for carrying out any particular simulation.
5. Details of how the conceptual model is expressed in a computable form, or how simulation results are to be recorded or analyzed, also depend on the hypothesis.

Which is not to say that you have to reveal any confidential ideas in the Forum. However, the utility of the answers that I can provide will depend on the degree to which I understand what you are trying to do. More than once I have seen users say "I want to do X" when it turned out that they really wanted to do Y, and that Y could be done in a way that was very easy. Case in point:

I would try to save the Voltage time curves for all neurons and then probably use some other software to detect when there was an action potential.

If spike times are the important result, just use the NetCon class's record() method to capture spike times to one or more Vectors. Much faster and consumes much less space than recording membrane potentials for post-simulation analysis.

I looked over the rest of the tutorial that I linked before.

Don't fall in love with that old tutorial. It contains some useful stuff, which is why we still link to it, but it's not the acme exemplar of all time.

I am wondering if through NMODL you are basically defining your own variables and section(which you can then "insert" to another section. . . . I am still curious though as to what you think I should read on if I wanted to try to do all of the things I listed before with just the hoc and NMODL languages?

Truly you are off on a tangent with this NMODL stuff. What you need to do is formulate a very clear conceptual model, based on a very clear hypothesis. Focus entirely on the biological question. The conceptual model should specify what kind of anatomical and biophysical features your cells should have.

You'll also want to consider what kinds of experiment(s) you would like to perform on the conceptual model. If the experiments involve drug effects, or effects of toxic agents, you need to specify how the drugs or toxins will affect model parameters, and whether those effects are to be represented mechanistically, or whether it will be sufficient to represent them phenomenologically.

[kgeorgiadi]

Hello again,

Thank you for the reply. I realize now I need to be more specific with my questions. After meeting with my neuroscientist supervisor we decided that the first step is for us to find an appropriate network from https://senselab.med.yale.edu/MicroCirc ... =19&lin=-1

However, the one which he suggested I take a look at (https://senselab.med.yale.edu/MicroCirc ... del=136095) has a lot of code that is hard to read(and very long) and I think that they don't use the normal HH and passive channels that NEURON has, which makes me unsure about which parts of the code are relevant and required simply for the network creation.

In any case, it would help tremendously if you were aware of a network available that has at least a couple(preferably more) of "small world networks" and some projection neurons connecting the small world networks. That would be the bare minimum. If it is also relatively easy code to read and has excitatory/inhibitory neurons and its typical output(without the pathogenic protein) is easily interpretable those would be big pluses.

[ted]

Bill Lytton's group does some interesting modeling. You might want to contact them directly for specific questions about their models.

In any case, it would help tremendously if you were aware of a network available that has at least a couple(preferably more) of "small world networks" and some projection neurons connecting the small world networks.

Nope.

If it is also relatively easy code to read and has excitatory/inhibitory neurons and its typical output(without the pathogenic protein) is easily interpretable those would be big pluses.

Nope again.

Which is not to say that such models do not exist or, if they do, are not readily available from ModelDB or some other source. Most network models with which I am even remotely familiar address a particular subset of features of a specific cortical or subcortical region, not whether they involve small world subnets.