Biological neuron models and simulated data

[uetmathematics]

I have some very simple questions.

1. What are the real life applications of the biological neuron models (for example Hodgkin-Huxely model)?

2. Is their any online database from where I can get the original recordings data of a neuron?

3. or is there any way that I can generate/simulate neuron data like real neuron?

 

[KCC2]

There are software for neuron simulations. One example: http://www.neuron.yale.edu/neuron/
I have never used them so I don't know how do they work and how much programming skills you need, but there might be tutorials/info on web.

 

[atyy]

1. What are the real life applications of the biological neuron models (for example Hodgkin-Huxely model)?

This is just my view, but I mainly think of the HH model as a synthesis of many things, including the need for concentrations of various ions in the extracellular fluid for correct neural functioning. So it and the knowledge it synthesizes is the basis of thing like knowing that some neural disorders can be treated rather simply by an intravenous infusion.

The HH model is very common in research, but I'm guessing that's not what you meant by real life. Also, for models of the brain, simpler models that capture less detail than the HH model are usually used. The idea is that to fit a HH model we need a lot of data, and for the brain regions we don't know so much, so if the model depended sensitively on some parameter, then maybe it's not so reliable.

2. Is their any online database from where I can get the original recordings data of a neuron?

http://incf.org/community/competitions/archive/spike-time-prediction/2007
http://incf.org/community/competitions/archive/spike-time-prediction/2008
http://incf.org/community/competitions/archive/spike-time-prediction/2009

3. or is there any way that I can generate/simulate neuron data like real neuron?

http://www.briansimulator.org/ free simulation software which is quite friendly.

 

[Pythagorean]

I've always been told that pharamecutical companies will hire computational neuroscientists but I've never actually seen such a job offer that mentions spiking models.

 

[Q_Goest]

The two main neuron simulation tools I've seen are called "Neuron" (KCC2 references this one) and "Genesis". Genesis seems to be open source or at least you can download off the internet. There is also a fairly extensive book that is freely available on the net that goes through the software and gives quite a bit of technical background as well.

I've not used any of these software programs before but I've looked into them and how they work. I'd be interested in any opinions people have of them.

 

[uetmathematics]

This is just my view, but I mainly think of the HH model as a synthesis of many things, including the need for concentrations of various ions in the extracellular fluid for correct neural functioning. So it and the knowledge it synthesizes is the basis of thing like knowing that some neural disorders can be treated rather simply by an intravenous infusion.

The HH model is very common in research, but I'm guessing that's not what you meant by real life. Also, for models of the brain, simpler models that capture less detail than the HH model are usually used. The idea is that to fit a HH model we need a lot of data, and for the brain regions we don't know so much, so if the model depended sensitively on some parameter, then maybe it's not so reliable.



http://incf.org/community/competitions/archive/spike-time-prediction/2007
http://incf.org/community/competitions/archive/spike-time-prediction/2008
http://incf.org/community/competitions/archive/spike-time-prediction/2009



http://www.briansimulator.org/ free simulation software which is quite friendly.

Thank you very much for the reply.
I want to know that what is the possible use of these models in any kind of application for instance brain-computer interface? is there any such application developed yet?

Thank you all for replying my post.

 

[atyy]

Thank you very much for the reply.
I want to know that what is the possible use of these models in any kind of application for instance brain-computer interface? is there any such application developed yet?

Usually in brain computer interfaces, the relationship between the intended action and the spiking activity of the brain is complicated, and the underlying connectivity is not exactly known. I'm not aware of any brain-computer interfaces which use Hodgkin-Huxley neurons. The dynamical system model in a brain-computer interface usually doesn't have such a direct interpretation in terms of the activities of specific neurons.

HH models have been used to try to understand the mechanisms of deep-brain stimulation for Parkinson's. http://www.ncbi.nlm.nih.gov/pubmed/14668299

 

[jbrussell93]

SNNAP is a very user friendly and straight forward neurosimulator that requires absolutely no programming with a decent GUI. You can simulate a single neuron or even networks with considerable customization. It is free to download.
http://nba.uth.tmc.edu/snnap/

Also, you have the option to use Hodgkin Huxley

I have used this program extensively in the past so let me know if you have any questions about getting started.

As for the application, I was working on a project studying larval lamprey neurons a couple of years back. It turns out they are able to fully regenerate their spinal cord a few weeks after being snipped nearly in half! We suspected that the calcium channels were probably playing a large role in this regeneration. Experiments were done to get actual action potentials of damaged neurons and I used SNNAP to try and model the action potential by modifying the ion channels. This would hopefully give us an idea of what ion channels are active, or not, in the damaged neurons.

I ended up leaving the lab with few results most likely due to the simplicity of the program. I was unable to model the damaged neuron action potential using realistic parameters. Great learning experience though! ;)

 

[uetmathematics]

SNNAP is a very user friendly and straight forward neurosimulator that requires absolutely no programming with a decent GUI. You can simulate a single neuron or even networks with considerable customization. It is free to download.
http://nba.uth.tmc.edu/snnap/

Also, you have the option to use Hodgkin Huxley

I have used this program extensively in the past so let me know if you have any questions about getting started.

As for the application, I was working on a project studying larval lamprey neurons a couple of years back. It turns out they are able to fully regenerate their spinal cord a few weeks after being snipped nearly in half! We suspected that the calcium channels were probably playing a large role in this regeneration. Experiments were done to get actual action potentials of damaged neurons and I used SNNAP to try and model the action potential by modifying the ion channels. This would hopefully give us an idea of what ion channels are active, or not, in the damaged neurons.

I ended up leaving the lab with few results most likely due to the simplicity of the program. I was unable to model the damaged neuron action potential using realistic parameters. Great learning experience though! ;)

thank you very much for the detailed reply :)
Can i get single neurons recordings? is there any database available?
I have downloaded SNNAP, but how can i do my own simulation?
How can i get data series using SNNAP?

 

[jbrussell93]

Can i get single neurons recordings? is there any database available?

I'm not sure about a database exactly, but there are plenty of images of action potentials on google. You might even search google scholar for the old Hodgkin-Huxley papers (or similar) to see some real recordings.

I have downloaded SNNAP, but how can i do my own simulation?
How can i get data series using SNNAP?

The first thing to note is that you should run SNNAP from the desktop, otherwise you will get "javascript" errors when you try to run the simulations. I would just put the entire folder you downloaded to the desktop.

I would suggest playing with some of the examples first. You can even build off of the examples, that is what I did. Try this one to start: From the main menu click "Run Simulation" then from simulation screen click at the top left...
"File > Load simulation > [snnap file destination you chose] > Examples > HH_type_neurons > Biophysics_01 > Lab_01_squid > Spike > Spike_05.smu"
(.smu files are simulation files)

Once it is loaded, click the "start" button

The white graph on top displays the action potential (voltage of cell with time) and the red/blue graph below shows the respective Na/K current with time! Pretty cool huh... It's governed completely by the HH equations.

The best way to understand how this program works and how you can customize is by clicking "Edit Formula" from the main screen. This will show you the tree of how the model is built. It starts with the trunk of the tree on the left with ntw (network) and branches out to neu (neuron) where you can edit which ion channels are active and how they communicate. From there it branches out to the equations/parameters governing the ion channels which might be a bit too detailed for your purposes.

It takes some time to figure out how everything connects together and it can be a bit quirky but it's actually pretty fun once you figure it out. Enjoy!

 

[Pythagorean]

I'd suggest the Collaborative Research in Computational Neuroscience database:

http://crcns.org/

It's where I've found recordings from rat neurons before to tune my models with a genetic algorithm. To return to an older question of yours, for modelling, I prefer to use a general high-level programming language like MATLAB or Python and write up the neuron model myself as a set of differential equations.

 

[uetmathematics]

I'm not sure about a database exactly, but there are plenty of images of action potentials on google. You might even search google scholar for the old Hodgkin-Huxley papers (or similar) to see some real recordings.


The first thing to note is that you should run SNNAP from the desktop, otherwise you will get "javascript" errors when you try to run the simulations. I would just put the entire folder you downloaded to the desktop.

I would suggest playing with some of the examples first. You can even build off of the examples, that is what I did. Try this one to start: From the main menu click "Run Simulation" then from simulation screen click at the top left...
"File > Load simulation > [snnap file destination you chose] > Examples > HH_type_neurons > Biophysics_01 > Lab_01_squid > Spike > Spike_05.smu"
(.smu files are simulation files)

Once it is loaded, click the "start" button

The white graph on top displays the action potential (voltage of cell with time) and the red/blue graph below shows the respective Na/K current with time! Pretty cool huh... It's governed completely by the HH equations.

The best way to understand how this program works and how you can customize is by clicking "Edit Formula" from the main screen. This will show you the tree of how the model is built. It starts with the trunk of the tree on the left with ntw (network) and branches out to neu (neuron) where you can edit which ion channels are active and how they communicate. From there it branches out to the equations/parameters governing the ion channels which might be a bit too detailed for your purposes.

It takes some time to figure out how everything connects together and it can be a bit quirky but it's actually pretty fun once you figure it out. Enjoy!

Yes, you are right. the software is quite easy and user friendly. that is what i am looking for thanks to you :)
I have tried the first part by loading and starting the simulations based on the examples.
also i have tried to edit those examples.
but i could not find where can i can see the values of membrane potential and ion channels at each time instant. where can i find this?

 

[gravenewworld]

So, since I'm completely unfamiliar with modeling neurons, I have a few questions.

One thing that I've always wondered is that with all of the modeling that is going on with neurons and ion channels, does anyone ever take into account ion channel glycosylation? It has been said before, that pretty much for almost any ion channel 30% of the entire molecular weight comes from glycan structures on the protein. Yet, whenever I see news blurbs about X, Y, or Z model of an ion channel, I see no mention of ion channel glycosylation at all, and probably 99% of the time, no depiction of glycan structures on ion channels in cartoons that are supposed to represent an ion channel. Who cares about glycosylation? Well, for one, the way an ion channel is glycosylated significantly alters its electrical properties and action potentials,and is absolutely critical for correct ion channel function:

http://www.ncbi.nlm.nih.gov/pubmed/22493431
http://jp.physoc.org/content/589/19/4647
http://www.ncbi.nlm.nih.gov/pubmed/23798301

Also, neuronal plasticity can fundamentally be altered by things like polymers of sialic acid that get attached to Neural Cell Adhesion Molecules (NCAMs), which are believed to play very important roles in learning, memory, and neuron plasticity.

http://www.sciencedirect.com/science/article/pii/089662739090310CNow my question is, how do these models account for this type of biology (if at all)? If it isn't in any of these models, what or how much limitation does this impose? Many of the above forms of glycosylation have either no modeling or very poor models to describe them because there is no known template to predict when, where, how, or even what type of glycosylation will happen. Glycans are also constantly being shed and remodeled dynamically to respond to environment and modulate biology, which in this case could be the way neurons fire or how they organize. If you can not model this, can you include some corrective factor as a catch all to include uncertainty in models of neurons?

 

[Pythagorean]

Modeling modulators and second messenger systems usually means that your Boltzmann distributions (activation functions) change a bit (as a function of the concentration of said second messenger).

You can also explicitly model second messenger interactions with mass action kinetics or Michaelis Menten kinetics. It's often necessary for calcium based currents, since calcium acts on ion channels.

I don't know anything about glycosylation, but I imagine where it plays a significant role it could be treated similarly.

 

[gravenewworld]

Modeling modulators and second messenger systems usually means that your Boltzmann distributions (activation functions) change a bit (as a function of the concentration of said second messenger).

You can also explicitly model second messenger interactions with mass action kinetics or Michaelis Menten kinetics. It's often necessary for calcium based currents, since calcium acts on ion channels.

I don't know anything about glycosylation, but I imagine where it plays a significant role it could be treated similarly.

So one of the other things that is interesting though is that glycosylation not only affects ion channel electrical activity, but also its stability and cell surface trafficking, for example GABA :

http://onlinelibrary.wiley.com/doi/10.1111/j.1742-4658.2005.04595.x/full

or sodium ion channels:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3757325/


I assume receptor density and total number is important in models of the neuron? Can models of the neuron and neural networks take changes in the total number ion channels into account for individual neurons since in vivo that is likely to be the case? Neurons are most likely constantly changing the way ion channels are being glycosylated to not only fine tune electrical activity across their membranes, but work to control even the number of channels that appear on the surface and total individual neuron electrical properties. I imagine in real life neuron X may say have 100 ion channels on its surface (just a hypothetical number to illustrate the idea) that are glycosylated in a certain way to produce a desired electrical response while neuron Y that is 200 neurons away my say have only 75 ion channels on its surface that are glycosylated completely differently at that same exact moment in time. The point is that in vivo, there is probably very fine micro tuning of electrical signals being sent through networks of neurons, where each neuron could behave as an individual by changing glycosylation patterns on receptors which can not only change electrical response, but also total # of receptors, and even where the receptors can be localized. So basically I guess you could think of it like youhad a population of neurons with all different properties. How could this be included in your neural network model? Maybe someone has already done this?

 

[Pythagorean]

Changing the number of channels would just change a constant in the model: the maximum conductivity for that channel, since each channel is assumed to have the same conductance.

To what extent could the mechanisms you're thinking of be regulatory: in that their purpose is to keep the neuron functioning normally? (To compensate for o ther changes in the cell).

Often, with good data, we are able to fit a model that reproduces the experimental neuron's behavior without considering such things.

I'm currently collaborating with an experimenter and I can reproduce the spiking behavior for individual neurons by just fitting the standard Hodgkin Huxley kinetics.

 

[Pythagorean]

Another thought: experimentalists are usually only reporting neurons that produce robust results. So if a neuron is varying a bunch (and not as a mechanism of something they're studying) then they won't use that neuron. When experimentalists do produce robust kinetics (which is generally what gets published) they match the basic assumptions of Hodgkin-Huxley quite well: a single cell withstands several voltage clamp treatments, maintaining a robust response in terms of activation/inactivation parameters. And these activation and inactivation functions allow the model to reproduce the output given a particular input, confirming the empirical physical assumptions about ion flow and electrochemistry across a capacitive membrane with a Boltzmann distribution of channel sensitivities.

Many (most) cells, of course, receive some modulatory effects and interested experimentalists will isolate those and characterize them and characterize their effects on the kinetics as a function of concentration in some microdomain near the channel (like calcium channels are often near other channels with calcium targets like calmodulin).

Then, if that's the phenomena the modeler is interested in, they will try to fit their activation constants as functions of concentration and introduce a new variable that models the concentration as a function of activity. If the concentration changes come from a completely different system that you're not interested in studying, and has no functional dependence on the state of the neuron you're studying, then you just model it in an algorithmic matter (there's really nothing else you can do).

If you're actually interested in the mechanics of the protein system, you're getting more into modeling proteomic and genomic dynamics than neuron dynamics. As a neuron modeler, I might draw on models from proteomics if I think it's necessary, but I wouldn't do all the tinkering to model a proteomic system myself.

There's actually a book called "Computational Neurogenetics" that has treatments whereby you consider networks of neurons, but each neuron in the network is really a network of genetic/protein processes. So you have a network of networks:

http://www.springer.com/engineering/biomedical+engineering/book/978-0-387-48353-5

 

[gravenewworld]

Changing the number of channels would just change a constant in the model: the maximum conductivity for that channel, since each channel is assumed to have the same conductance.

To what extent could the mechanisms you're thinking of be regulatory: in that their purpose is to keep the neuron functioning normally? (To compensate for o ther changes in the cell).

Often, with good data, we are able to fit a model that reproduces the experimental neuron's behavior without considering such things.

I'm currently collaborating with an experimenter and I can reproduce the spiking behavior for individual neurons by just fitting the standard Hodgkin Huxley kinetics.

That's my point I guess, is that there are explainable mechanisms through which each channel in a neuron could have different conductance and through which the number of channels on the same neuron can change with time based on the way it is modified with glycans. It's almost impossible to predict how an ion channel would be glycosylated at a moment in time. Could the H-H include a component so that the overall electrical activity of a single neuron changes with time? When you study the H-H model, what kind of environment is your neuron in? Can you test it with say one concentration of insulin and then also test it with say a different concentration of insulin? Could you test a neuron in different concentrations of sugars? Then check to see if electrical activity has a measurable change?

Now what happens when you connect a billion neurons together? Where each neuron has its own electrical identity. This identity of each neuron could also change in time, because something like glycosylation responds to environmental stress and cue. How can one include the fact that each neuron in a network is just slightly different and the identity of each is constantly changing? What I'm getting at is, is it possible to take into account the dynamic nature of an in vivo system? Hormonal cues are constantly changing, responding to stress, and responding differently in different parts of the body. It could be another way to fine tune electrical activity through a network of neurons so that the behavior of different parts of a neural network are different. Who knows, maybe it could be important for learning or storing memory? I have no idea for what other purposes dynamically changing regulatory mechanisms that fine tune electrical activity of ion channels(like glycosylation) may be important for, but nature is likely doing it for a reason.

 

[gravenewworld]

Another thought: experimentalists are usually only reporting neurons that produce robust results. So if a neuron is varying a bunch (and not as a mechanism of something they're studying) then they won't use that neuron. When experimentalists do produce robust kinetics (which is generally what gets published) they match the basic assumptions of Hodgkin-Huxley quite well: a single cell withstands several voltage clamp treatments, maintaining a robust response in terms of activation/inactivation parameters. And these activation and inactivation functions allow the model to reproduce the output given a particular input, confirming the empirical physical assumptions about ion flow and electrochemistry across a capacitive membrane with a Boltzmann distribution of channel sensitivities.

Many (most) cells, of course, receive some modulatory effects and interested experimentalists will isolate those and characterize them and characterize their effects on the kinetics as a function of concentration in some microdomain near the channel (like calcium channels are often near other channels with calcium targets like calmodulin).

Then, if that's the phenomena the modeler is interested in, they will try to fit their activation constants as functions of concentration and introduce a new variable that models the concentration as a function of activity. If the concentration changes come from a completely different system that you're not interested in studying, and has no functional dependence on the state of the neuron you're studying, then you just model it in an algorithmic matter (there's really nothing else you can do).

If you're actually interested in the mechanics of the protein system, you're getting more into modeling proteomic and genomic dynamics than neuron dynamics. As a neuron modeler, I might draw on models from proteomics if I think it's necessary, but I wouldn't do all the tinkering to model a proteomic system myself.

There's actually a book called "Computational Neurogenetics" that has treatments whereby you consider networks of neurons, but each neuron in the network is really a network of genetic/protein processes. So you have a network of networks:

http://www.springer.com/engineering/biomedical+engineering/book/978-0-387-48353-5

Hmm interesting, I may really check this book out. Thanks!

 

[Pythagorean]

That's my point I guess, is that there are explainable mechanisms through which each channel in a neuron could have different conductance and through which the number of channels on the same neuron can change with time based on the way it is modified with glycans. It's almost impossible to predict how an ion channel would be glycosylated at a moment in time. Could the H-H include a component so that the overall electrical activity of a single neuron changes with time? When you study the H-H model, what kind of environment is your neuron in? Can you test it with say one concentration of insulin and then also test it with say a different concentration of insulin? Could you test a neuron in different concentrations of sugars? Then check to see if electrical activity has a measurable change?

Now what happens when you connect a billion neurons together? Where each neuron has its own electrical identity. This identity of each neuron could also change in time, because something like glycosylation responds to environmental stress and cue. How can one include the fact that each neuron in a network is just slightly different and the identity of each is constantly changing? What I'm getting at is, is it possible to take into account the dynamic nature of an in vivo system? Hormonal cues are constantly changing, responding to stress, and responding differently in different parts of the body. It could be another way to fine tune electrical activity through a network of neurons so that the behavior of different parts of a neural network are different. Who knows, maybe it could be important for learning or storing memory? I have no idea for what other purposes dynamically changing regulatory mechanisms that fine tune electrical activity of ion channels(like glycosylation) may be important for, but nature is likely doing it for a reason.

So channels are already considered stochastically. The activation function is based on a maxwell Boltzmann distribution of channel sensitivities. If glyocsylation was pervasive (I have no idea of how common it is) it could be a part of that distribution.

So we can't (and don't try) to predict individual channels deterministically, but we predict the group behavior of the channels as a statistical ensemble. The processes that go into determining each channel's individual state could have glycosylation involved for all I know.

People do distinguish heterogenous networks form homogenous ones (all the same neuron in a network vs. a distribution of neurons) and can find different dynamics in each case. That's a common topic in complexity sciences in many models besides neurons.

When you study the H-H model, what kind of environment is your neuron in? Can you test it with say one concentration of insulin and then also test it with say a different concentration of insulin? Could you test a neuron in different concentrations of sugars? Then check to see if electrical activity has a measurable change?

This would be a similar extension to the model, but the canonical model doesn't contain that machinery. You'd need experimental data to augment the Hodgkin Huxley model. As above, the experimentalist would have to measure the effects on the kinetics of the electrophysiology and then the modeler would fit that data to a function that replaces a constant in the HH model.

 

[gravenewworld]

So channels are already considered stochastically. The activation function is based on a maxwell Boltzmann distribution of channel sensitivities. If glyocsylation was pervasive (I have no idea of how common it is) it could be a part of that distribution.

So we can't (and don't try) to predict individual channels deterministically, but we predict the group behavior of the channels as a statistical ensemble. The processes that go into determining each channel's individual state could have glycosylation involved for all I know.

People do distinguish heterogenous networks form homogenous ones (all the same neuron in a network vs. a distribution of neurons) and can find different dynamics in each case. That's a common topic in complexity sciences in many models besides neurons.
This would be a similar extension to the model, but the canonical model doesn't contain that machinery. You'd need experimental data to augment the Hodgkin Huxley model. As above, the experimentalist would have to measure the effects on the kinetics of the electrophysiology and then the modeler would fit that data to a function that replaces a constant in the HH model.

So what would be the advantages of having subceullar prediction for ion channel activity? Instead of using stochastic models, what would you need to have a deterministic model?

Glycosylation is everywhere, in every single cell. If you look at the glycome of the heart--that is the entire network of glycosylation enzymes that are expressed and all of the patterns of glycans that they can produce---there are different but non-random patterns of glycans on ion channels depending on which portion you are in of the heart. People believe that the glycome is tuning patterns glycosylation on ion channels in a non-random manner to tightly control electrical current through heart tissue. There's likely a high probability that the brain is the same.

I sort of remember stochastic processes from my undergrad when we went over brownian motion, which made the assumption of random walks to be able to apply stochastic models to describe motion of the particles. What if in this case ion channel activity was not random based on glycosylation of ion channels, that is inherently linked to a cell's metabolic state? A neuron may be non-randomly selecting for a specific population of ion channels to tune electrical response just like heart tissues seem to be doing.

I'm assuming that HH experiments are done in tightly controlled conditions? How big are the gaps between in vitro HH experiments and how neurons behave in vivo in a dynamically changing environments?

 

[Pythagorean]

I don't know the details of glyosylation interactions, but most chemical systems have models based on mass action. But other dynamics are used, depending on the cooperativity or competitiveness of all the participating molecules. The variables/dimensions of the system are the concentrations of each participating molecule. A generic model of complex chemical interactions is the Grey Scott model:

http://groups.csail.mit.edu/mac/projects/amorphous/GrayScott/

The advantages/disadvantages depend on the specific question you're asking. Because of that previously mentioned problem of generality vs. specificity in modelling, you want to build your models to answer specific questions and not try to make a catch all. So such a model would be advantageous in predicting how glyosylation contributes to the kinetic state of individual channels the way you're framing it currently.

If you want to ask about whether the high frequency gamma oscillations in the hippocampus can be explained by axo-axonic gap junctions, you probably don't want to bother with the details of glycosylation.

I'm assuming that HH experiments are done in tightly controlled conditions? How big are the gaps between in vitro HH experiments and how neurons behave in vivo in a dynamically changing environments?

Yes, HH-type modeling is based off intracellular recordings. In Vivo experiments are more likely to be extracellular: measuring field potentials. Dynamically changing environments are modeled as "bath applications" or "volume transmission". It's basically just modeling more channels (extra-synaptic ones) and again, adding a new variable that represents the concentration of some ligand in the region.

You could do spatial extensions too, so that your ligands can diffuse and can have sources and sinks in 3D space and your "spherical cow" neurons would have a spatial designation and each neuron in the region of ligand would have channels reacting to it. I don't know if anyone actually does this, but I have a framework for how I'd model it. This kind of complexity is a lot of pieces to put together though, lots of experiments, lots of model tweaking, lots of abstractions. It would be a chore and you never know whether you're going to get an answer out of it or not so you have to pick your battles and work your way up to this kind of complexity carefully.