{
  "id": 7554,
  "title": "Help with Intuition",
  "url": "/competitions/connectomics/discussion/7554",
  "author_name": "",
  "post_date": "2014-03-30T17:51:13.687Z",
  "votes": 1,
  "comment_count": 3,
  "views": 947,
  "content": "<p>Hi - I'm joining the party a little late... I've read the starter material and earlier posts and I have some background in biology so I understand the fundamental elements of the project (neural activity, etc.).&nbsp; I do have a few questions about how to interpret and model the data, though.&nbsp; Hoping one of you can shed some light.</p>\n<p>I'll use a sample scenario to hopefully make this make sense:</p>\n<p>Neuron A fires and then triggers neurons B and X.&nbsp; B triggers C and X triggers Y.&nbsp; Finally Y triggers Z.&nbsp; From a sequencing standpoint:</p>\n<p>A -&gt; B -&gt; C</p>\n<p>A -&gt; X -&gt; Y -&gt; Z</p>\n<p>Questions:</p>\n<p>- I understand the dataset period to be 20ms.&nbsp; How fast do neurons fire?&nbsp; Is it reasonable to assume all of the interaction above will happen in one period?&nbsp; Two?&nbsp; Four?&nbsp; More?</p>\n<p>- In the example above, I expect a correlation analysis would show a relationship between X and C or B and Y (for example) even though physically there is none.&nbsp; So I assume the model has to be able to filter out firing events that have no causal relationhip even though the timing may show a potential connection?</p>\n\n<p>I'm sure there's more but I'll stop there for now.&nbsp; Thanks for your input.</p>\n\n<p>Cheers,</p>\n<p>Dan</p>",
  "messages": [
    {
      "id": "41197",
      "postDate": "03/30/2014 17:51:13",
      "content": "<p>Hi - I'm joining the party a little late... I've read the starter material and earlier posts and I have some background in biology so I understand the fundamental elements of the project (neural activity, etc.).&nbsp; I do have a few questions about how to interpret and model the data, though.&nbsp; Hoping one of you can shed some light.</p>\n<p>I'll use a sample scenario to hopefully make this make sense:</p>\n<p>Neuron A fires and then triggers neurons B and X.&nbsp; B triggers C and X triggers Y.&nbsp; Finally Y triggers Z.&nbsp; From a sequencing standpoint:</p>\n<p>A -&gt; B -&gt; C</p>\n<p>A -&gt; X -&gt; Y -&gt; Z</p>\n<p>Questions:</p>\n<p>- I understand the dataset period to be 20ms.&nbsp; How fast do neurons fire?&nbsp; Is it reasonable to assume all of the interaction above will happen in one period?&nbsp; Two?&nbsp; Four?&nbsp; More?</p>\n<p>- In the example above, I expect a correlation analysis would show a relationship between X and C or B and Y (for example) even though physically there is none.&nbsp; So I assume the model has to be able to filter out firing events that have no causal relationhip even though the timing may show a potential connection?</p>\n\n<p>I'm sure there's more but I'll stop there for now.&nbsp; Thanks for your input.</p>\n\n<p>Cheers,</p>\n<p>Dan</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "41224",
      "postDate": "03/31/2014 13:31:01",
      "content": "<p>Hi Dan,</p>\n<p>Welcome to the party!</p>\n<p>1) Yes, sampling interval is 20ms, and neurons can fire multiple times in that time window. Current influence/causality detection methods don't look at time signals more than 3 frames apart, you can take that as a first estimate, but you should explore further away, who knows!&nbsp;</p>\n<p>2) Keep in mind that the system shows two regimes of activity (low and high). On the low activity regime it is very unlikely that something like your example would happen. The firing rate in that regime is very low, and the probability that a firing in A triggers a firing in B is very small. The neuron requires quite a few simultaneous inputs to certainly fire. On the high activity regime however, it might seem that everything causes everything, since all the traces seem to go up in the same frame. Some reconstruction methods avoid this high activity regime for example. Although there's interesting information there too !</p>\n\n<p>Best,</p>\n<p>Javier</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "41249",
      "postDate": "03/31/2014 18:32:21",
      "content": "<p>Thanks Javier.&nbsp; So, based on your response above, from a biological perspective I think you're saying that the following is the most likely pattern during 'low' activity:</p>\n<p>A B and C all fire (perhaps based on some unmeasured stimulus).&nbsp; 1-3 time periods later X fires.&nbsp; One can then start building a case that A, B and C all have a directed linkage to X.</p>\n<p>Is there a biological case for assuming a directed/causal relationship between A B and C as well?</p>\n\n<p>Re. distance, is there a biological basis for weighting the potential connection between two neurons by the inverse of the distance?&nbsp; Or does distance not play into the causality?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "41390",
      "postDate": "04/02/2014 11:59:46",
      "content": "<p>1) Yes. However, most of the time X would fire on the same period as the others.</p>\n<p>2) From the fact that X fired after A,B,C? Not really. That's directly related to the clustering coefficient of the network, and you don't know if its's low or high beforehand.</p>\n<p>3) Distance plays no role in causality in this challenge.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 41224,
      "author_name": "orlandi",
      "author_url": "",
      "post_date": "03/31/2014 13:31:01",
      "content": "<p>Hi Dan,</p>\n<p>Welcome to the party!</p>\n<p>1) Yes, sampling interval is 20ms, and neurons can fire multiple times in that time window. Current influence/causality detection methods don't look at time signals more than 3 frames apart, you can take that as a first estimate, but you should explore further away, who knows!&nbsp;</p>\n<p>2) Keep in mind that the system shows two regimes of activity (low and high). On the low activity regime it is very unlikely that something like your example would happen. The firing rate in that regime is very low, and the probability that a firing in A triggers a firing in B is very small. The neuron requires quite a few simultaneous inputs to certainly fire. On the high activity regime however, it might seem that everything causes everything, since all the traces seem to go up in the same frame. Some reconstruction methods avoid this high activity regime for example. Although there's interesting information there too !</p>\n\n<p>Best,</p>\n<p>Javier</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 41249,
      "author_name": "ecodan",
      "author_url": "",
      "post_date": "03/31/2014 18:32:21",
      "content": "<p>Thanks Javier.&nbsp; So, based on your response above, from a biological perspective I think you're saying that the following is the most likely pattern during 'low' activity:</p>\n<p>A B and C all fire (perhaps based on some unmeasured stimulus).&nbsp; 1-3 time periods later X fires.&nbsp; One can then start building a case that A, B and C all have a directed linkage to X.</p>\n<p>Is there a biological case for assuming a directed/causal relationship between A B and C as well?</p>\n\n<p>Re. distance, is there a biological basis for weighting the potential connection between two neurons by the inverse of the distance?&nbsp; Or does distance not play into the causality?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 41390,
      "author_name": "orlandi",
      "author_url": "",
      "post_date": "04/02/2014 11:59:46",
      "content": "<p>1) Yes. However, most of the time X would fire on the same period as the others.</p>\n<p>2) From the fact that X fired after A,B,C? Not really. That's directly related to the clustering coefficient of the network, and you don't know if its's low or high beforehand.</p>\n<p>3) Distance plays no role in causality in this challenge.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "41197": "",
    "41224": "",
    "41249": "",
    "41390": ""
  },
  "source": "meta"
}