{
  "id": 7518,
  "title": "Neuron location and connection probability",
  "url": "/competitions/connectomics/discussion/7518",
  "author_name": "",
  "post_date": "2014-03-26T00:09:01.690Z",
  "votes": null,
  "comment_count": 5,
  "views": 1289,
  "content": "<p>Hi all,</p>\n<p>I have a question for the other participants as well as possibly the competition organizers. &nbsp;In the extended description of the datasets (https://sites.google.com/a/chalearn.org/connectomics/help/data), the organizers say that, in modeling neuronal connectivity, they use &quot;topological models: connectivity models taking into account neuron proximity and enforcing a realistic degree of node clustering&quot;. &nbsp;However, based on some preliminary investigation, I've been unable to find a relation between the probability of connection between two neurons and their spatial proximity (naively, one would expect nearby neurons to be more likely to be connected). &nbsp;Has anyone found a similar lack of relationship?</p>\n<p>Thanks!</p>\n<p>-a</p>",
  "messages": [
    {
      "id": "41021",
      "postDate": "03/26/2014 00:09:01",
      "content": "<p>Hi all,</p>\n<p>I have a question for the other participants as well as possibly the competition organizers. &nbsp;In the extended description of the datasets (https://sites.google.com/a/chalearn.org/connectomics/help/data), the organizers say that, in modeling neuronal connectivity, they use &quot;topological models: connectivity models taking into account neuron proximity and enforcing a realistic degree of node clustering&quot;. &nbsp;However, based on some preliminary investigation, I've been unable to find a relation between the probability of connection between two neurons and their spatial proximity (naively, one would expect nearby neurons to be more likely to be connected). &nbsp;Has anyone found a similar lack of relationship?</p>\n<p>Thanks!</p>\n<p>-a</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "41042",
      "postDate": "03/26/2014 14:51:25",
      "content": "<p>I concur. &nbsp;I tried using distance as the only&nbsp;measure of connectivity, and my results for all of the&nbsp;training sets were .5. &nbsp;Since .5 is also the result for &quot;random&quot;, I took this result to mean that distance and connectivity lacked any sort of relationship.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "41051",
      "postDate": "03/26/2014 16:53:55",
      "content": "<p>Hi,</p>\n<p>Actually that text is a little misleading, since it refers to a more general framework. In the current datasets there's no relationship between connection probability and spatial proximity. However, the positions do affect the fluorescence signal (due to light scattering effects).</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "43179",
      "postDate": "04/27/2014 18:48:55",
      "content": "<p>Hi Javier,</p>\n<p>Just to clarify.</p>\n<p>Are you implying the positions data do not help with our predictions because it wasn't included in the simulation model that generated all the datasets? Meaning, predictions could get worse if we use the positions data but it won't do better?</p>\n<p>Or, is it the case that the positions data could potentially&nbsp;help with addressing some noise issues arising from light scattering from nearby neurons?</p>\n\n<p>James.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "43274",
      "postDate": "04/29/2014 08:29:50",
      "content": "<p>[quote=James Tee;43179]</p>\n<p>Or, is it the case that the positions data could potentially&nbsp;help with addressing some noise issues arising from light scattering from nearby neurons?</p>\n<p>[/quote]</p>\n<p>That's it.</p>\n\n<p>Best,</p>\n<p>Javier</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "43388",
      "postDate": "04/30/2014 19:44:17",
      "content": "<p>Actually, I&nbsp;have found a statistically significant correlation. The connection probability appears to be approximately exp{-2*r^s} where r is the distance. Since straight lines don't fit exponential declines very well, if you are using a regression based classifiers you must replace the distance with an exponential.</p>\n\n<p>Of course, this could be related to other things.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 41042,
      "author_name": "adm1444",
      "author_url": "",
      "post_date": "03/26/2014 14:51:25",
      "content": "<p>I concur. &nbsp;I tried using distance as the only&nbsp;measure of connectivity, and my results for all of the&nbsp;training sets were .5. &nbsp;Since .5 is also the result for &quot;random&quot;, I took this result to mean that distance and connectivity lacked any sort of relationship.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 41051,
      "author_name": "orlandi",
      "author_url": "",
      "post_date": "03/26/2014 16:53:55",
      "content": "<p>Hi,</p>\n<p>Actually that text is a little misleading, since it refers to a more general framework. In the current datasets there's no relationship between connection probability and spatial proximity. However, the positions do affect the fluorescence signal (due to light scattering effects).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 43179,
      "author_name": "jamestee1",
      "author_url": "",
      "post_date": "04/27/2014 18:48:55",
      "content": "<p>Hi Javier,</p>\n<p>Just to clarify.</p>\n<p>Are you implying the positions data do not help with our predictions because it wasn't included in the simulation model that generated all the datasets? Meaning, predictions could get worse if we use the positions data but it won't do better?</p>\n<p>Or, is it the case that the positions data could potentially&nbsp;help with addressing some noise issues arising from light scattering from nearby neurons?</p>\n\n<p>James.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 43274,
      "author_name": "orlandi",
      "author_url": "",
      "post_date": "04/29/2014 08:29:50",
      "content": "<p>[quote=James Tee;43179]</p>\n<p>Or, is it the case that the positions data could potentially&nbsp;help with addressing some noise issues arising from light scattering from nearby neurons?</p>\n<p>[/quote]</p>\n<p>That's it.</p>\n\n<p>Best,</p>\n<p>Javier</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 43388,
      "author_name": "philipstephens",
      "author_url": "",
      "post_date": "04/30/2014 19:44:17",
      "content": "<p>Actually, I&nbsp;have found a statistically significant correlation. The connection probability appears to be approximately exp{-2*r^s} where r is the distance. Since straight lines don't fit exponential declines very well, if you are using a regression based classifiers you must replace the distance with an exponential.</p>\n\n<p>Of course, this could be related to other things.&nbsp;</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "41021": "",
    "41042": "",
    "41051": "",
    "43179": "",
    "43274": "",
    "43388": ""
  },
  "source": "meta"
}