{ "id": 7036, "title": "Question about neurons", "url": "/competitions/connectomics/discussion/7036", "author_name": "", "post_date": "2014-02-10T15:09:23.773Z", "votes": null, "comment_count": 8, "views": 1818, "content": "

@Compet Admin:

\n

What do you expect to see in the data (when conducting the simulation) for any given couple of connected neurons?

", "messages": [ { "id": "38553", "postDate": "02/10/2014 15:09:23", "content": "

@Compet Admin:

\n

What do you expect to see in the data (when conducting the simulation) for any given couple of connected neurons?

", "rawMarkdown": "", "votes": null }, { "id": "38591", "postDate": "02/10/2014 20:20:39", "content": "

Hi all, good question. If we knew a simple answer the problem of network reconstruction would be simple.

\n

The firing of a presynaptic neuron (i.e. the source of directed links) is expected to affect the probability of firing of its postsynaptic neurons (i.e. the targets of the directed links originating from the presynaptic neuron). If the presynaptic neuron is excitatory, the postsynaptic probability of firing will rise. If the presynaptic neuron was inhibitory, the postsynaptic probability of firing would decrease, but for this first challenge only excitatory synapses are active (which can be achieved as well in an experimental setting by blocking synapses with inhibitory action through specific drugs).

\n

However, a neuron receives a very large number of inputs, and retrieving the small modulations of probability of firing due to a single presynaptic neuron remains a hard task.

\n

Furthermore, neuronal cultures display in vivo (and in silico) characteristic behaviors such as synchronous bursting. Such behaviors are collective and cannot be explained easily in terms of pairwise interactions.

", "rawMarkdown": "", "votes": null }, { "id": "38634", "postDate": "02/11/2014 15:31:43", "content": "

I would like to add a further clarification. Salviati told: "The firing of a presynaptic neuron is expected to affect the probability of firing of its postsynaptic neurons. If the presynaptic neuron is excitatory, the postsynaptic probability of firing will rise."

\n

This is certainly true! But our simulations are not probabilistic, are real dynamical systems simulations. At every time-step we integrate a system of hundredths of non-linear differential equations, it is not just tossing a dice per neuron to decide whether it will fire or not. We have a full-fledged biophysically realistic simulation behind our simulated data.

\n

That said, at the phenomenological level, Salviati's comment is true: the post-synaptic probability of firing indeed will rise, if a pre-synaptic neuron fires… But, the amount of which it will rise is changing, depending as well on what all the other pre-synaptic neurons are doing.

", "rawMarkdown": "", "votes": null }, { "id": "39034", "postDate": "02/18/2014 13:43:57", "content": "

[quote=Sagredo;38634]

\n

But our simulations are not probabilistic, are real dynamical systems simulations. At every time-step we integrate a system of hundredths of non-linear differential equations, it is not just tossing a dice per neuron to decide whether it will fire or not. We have a full-fledged biophysically realistic simulation behind our simulated data.

\n

[/quote]

\n

When you say your 'simulations are not probabilistic', does this mean you don't include any stochastic elements within the simulated neuron dynamics (e.g. to model channel noise or unreliable synaptic transmission), i.e. that each neuron uses a fully deterministic leaky-integrate-and-fire model? If so is the only noise added to the data that in the simulation of the imaging dynamics?

", "rawMarkdown": "", "votes": null }, { "id": "39040", "postDate": "02/18/2014 14:14:57", "content": "

The system is stochastic. What Sagredo means is that all the references made on the previous posts to probabilities are only used to help understand the causal effects between neurons, the simulation never calculates probabilities per se.

", "rawMarkdown": "", "votes": null }, { "id": "39188", "postDate": "02/21/2014 05:49:13", "content": "

beginner question here.  Are we to assume anything about a time delay between a presynaptic and postsynaptic neuron.  One of the files gives the positions of the neurons...does this help us determine the time it takes a signal to travel from one neuron to the next...?

", "rawMarkdown": "", "votes": null }, { "id": "39196", "postDate": "02/21/2014 12:27:39", "content": "

I doubt this would help, as all neurons can communicate back and forth several times within one time bin (fluorescence signals are sampled at 20ms, action potentials travel at 1-100 mm/ms, synaptic delay <1ms, all neurons within 1 mm2). 

", "rawMarkdown": "", "votes": null }, { "id": "39197", "postDate": "02/21/2014 12:48:22", "content": "

Exactly.

\n

The positions files are provided mostly for convenience, in case you want to visualize the network with the actual positions.

\n

The fluorescence signal contains light scattering noise, which depend on the network positions but have nothing to do with the actual dynamics. As an indication, none of the sample code or benchmarks rely on the positions file.

", "rawMarkdown": "", "votes": null }, { "id": "39214", "postDate": "02/21/2014 18:15:52", "content": "

ah ok, thank you!

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 38591, "author_name": "salviati1", "author_url": "", "post_date": "02/10/2014 20:20:39", "content": "

Hi all, good question. If we knew a simple answer the problem of network reconstruction would be simple.

\n

The firing of a presynaptic neuron (i.e. the source of directed links) is expected to affect the probability of firing of its postsynaptic neurons (i.e. the targets of the directed links originating from the presynaptic neuron). If the presynaptic neuron is excitatory, the postsynaptic probability of firing will rise. If the presynaptic neuron was inhibitory, the postsynaptic probability of firing would decrease, but for this first challenge only excitatory synapses are active (which can be achieved as well in an experimental setting by blocking synapses with inhibitory action through specific drugs).

\n

However, a neuron receives a very large number of inputs, and retrieving the small modulations of probability of firing due to a single presynaptic neuron remains a hard task.

\n

Furthermore, neuronal cultures display in vivo (and in silico) characteristic behaviors such as synchronous bursting. Such behaviors are collective and cannot be explained easily in terms of pairwise interactions.

", "votes": null, "replies": [] }, { "id": 38634, "author_name": "sagredo", "author_url": "", "post_date": "02/11/2014 15:31:43", "content": "

I would like to add a further clarification. Salviati told: "The firing of a presynaptic neuron is expected to affect the probability of firing of its postsynaptic neurons. If the presynaptic neuron is excitatory, the postsynaptic probability of firing will rise."

\n

This is certainly true! But our simulations are not probabilistic, are real dynamical systems simulations. At every time-step we integrate a system of hundredths of non-linear differential equations, it is not just tossing a dice per neuron to decide whether it will fire or not. We have a full-fledged biophysically realistic simulation behind our simulated data.

\n

That said, at the phenomenological level, Salviati's comment is true: the post-synaptic probability of firing indeed will rise, if a pre-synaptic neuron fires… But, the amount of which it will rise is changing, depending as well on what all the other pre-synaptic neurons are doing.

", "votes": null, "replies": [] }, { "id": 39034, "author_name": "mattgraham", "author_url": "", "post_date": "02/18/2014 13:43:57", "content": "

[quote=Sagredo;38634]

\n

But our simulations are not probabilistic, are real dynamical systems simulations. At every time-step we integrate a system of hundredths of non-linear differential equations, it is not just tossing a dice per neuron to decide whether it will fire or not. We have a full-fledged biophysically realistic simulation behind our simulated data.

\n

[/quote]

\n

When you say your 'simulations are not probabilistic', does this mean you don't include any stochastic elements within the simulated neuron dynamics (e.g. to model channel noise or unreliable synaptic transmission), i.e. that each neuron uses a fully deterministic leaky-integrate-and-fire model? If so is the only noise added to the data that in the simulation of the imaging dynamics?

", "votes": null, "replies": [] }, { "id": 39040, "author_name": "orlandi", "author_url": "", "post_date": "02/18/2014 14:14:57", "content": "

The system is stochastic. What Sagredo means is that all the references made on the previous posts to probabilities are only used to help understand the causal effects between neurons, the simulation never calculates probabilities per se.

", "votes": null, "replies": [] }, { "id": 39188, "author_name": "idontknowwhatimdoing0", "author_url": "", "post_date": "02/21/2014 05:49:13", "content": "

beginner question here.  Are we to assume anything about a time delay between a presynaptic and postsynaptic neuron.  One of the files gives the positions of the neurons...does this help us determine the time it takes a signal to travel from one neuron to the next...?

", "votes": null, "replies": [] }, { "id": 39196, "author_name": "jay670790", "author_url": "", "post_date": "02/21/2014 12:27:39", "content": "

I doubt this would help, as all neurons can communicate back and forth several times within one time bin (fluorescence signals are sampled at 20ms, action potentials travel at 1-100 mm/ms, synaptic delay <1ms, all neurons within 1 mm2). 

", "votes": null, "replies": [] }, { "id": 39197, "author_name": "orlandi", "author_url": "", "post_date": "02/21/2014 12:48:22", "content": "

Exactly.

\n

The positions files are provided mostly for convenience, in case you want to visualize the network with the actual positions.

\n

The fluorescence signal contains light scattering noise, which depend on the network positions but have nothing to do with the actual dynamics. As an indication, none of the sample code or benchmarks rely on the positions file.

", "votes": null, "replies": [] }, { "id": 39214, "author_name": "idontknowwhatimdoing0", "author_url": "", "post_date": "02/21/2014 18:15:52", "content": "

ah ok, thank you!

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "38553": "", "38591": "", "38634": "", "39034": "", "39040": "", "39188": "", "39196": "", "39197": "", "39214": "" }, "source": "meta" }