{
  "id": 96882,
  "title": "molecule representation",
  "url": "/competitions/champs-scalar-coupling/discussion/96882",
  "author_name": "",
  "post_date": "2019-06-24T08:19:19.808356700Z",
  "votes": 66,
  "comment_count": 8,
  "views": 0,
  "content": "<p><a href=\"https://github.com/SINGROUP/dscribe\">https://github.com/SINGROUP/dscribe</a></p>\n\n<ul>\n<li>Coulomb matrix</li>\n<li>Sine matrix</li>\n<li>Ewald matrix</li>\n<li>Atom-centered Symmetry Functions (ACSF)</li>\n<li>Smooth Overlap of Atomic Orbitals (SOAP)</li>\n<li>Many-body Tensor Representation (MBTR)</li>\n<li>Local Many-body Tensor Representation (LMBTR)</li>\n</ul>\n\n<p><a href=\"https://github.com/atomistic-machine-learning/schnetpack\">https://github.com/atomistic-machine-learning/schnetpack</a>\n - wACSF - weighted atom-centered symmetry functions </p>",
  "messages": [
    {
      "id": "559530",
      "postDate": "06/24/2019 08:19:19",
      "content": "<p><a href=\"https://github.com/SINGROUP/dscribe\">https://github.com/SINGROUP/dscribe</a></p>\n\n<ul>\n<li>Coulomb matrix</li>\n<li>Sine matrix</li>\n<li>Ewald matrix</li>\n<li>Atom-centered Symmetry Functions (ACSF)</li>\n<li>Smooth Overlap of Atomic Orbitals (SOAP)</li>\n<li>Many-body Tensor Representation (MBTR)</li>\n<li>Local Many-body Tensor Representation (LMBTR)</li>\n</ul>\n\n<p><a href=\"https://github.com/atomistic-machine-learning/schnetpack\">https://github.com/atomistic-machine-learning/schnetpack</a>\n - wACSF - weighted atom-centered symmetry functions </p>",
      "rawMarkdown": "https://github.com/SINGROUP/dscribe\n\n- Coulomb matrix\n- Sine matrix\n- Ewald matrix\n- Atom-centered Symmetry Functions (ACSF)\n- Smooth Overlap of Atomic Orbitals (SOAP)\n- Many-body Tensor Representation (MBTR)\n- Local Many-body Tensor Representation (LMBTR)\n\nhttps://github.com/atomistic-machine-learning/schnetpack\n - wACSF - weighted atom-centered symmetry functions",
      "votes": null
    },
    {
      "id": "560045",
      "postDate": "06/25/2019 01:36:53",
      "content": "<p>Get angle information from rdkit\n<a href=\"https://www.google.com/amp/s/iwatobipen.wordpress.com/2016/07/16/get-molecular-angle-information-using-rdkit/amp/\">https://www.google.com/amp/s/iwatobipen.wordpress.com/2016/07/16/get-molecular-angle-information-using-rdkit/amp/</a></p>",
      "rawMarkdown": "Get angle information from rdkit\nhttps://www.google.com/amp/s/iwatobipen.wordpress.com/2016/07/16/get-molecular-angle-information-using-rdkit/amp/",
      "votes": null
    },
    {
      "id": "560107",
      "postDate": "06/25/2019 03:56:26",
      "content": "<p>this is a very good thread heng. i want to add to my fellow competitors that the literature has many variations on molecule representation , and we dont know the best. one paper i read said that graph theory/connected edge is one of the best representations. another says look into representing distances between atoms, and also angles between them. Also i saw a paper convert some of the \"molecule\" representations into \"atomic\" representations by instead of considering all atoms in the molecule, only consider neighbor atoms within a Radius of the atoms you are focusing on. The literature seems shotgunned with so many different molecule techniques that it will be exciting to see the variation of representations between the winners</p>",
      "rawMarkdown": "this is a very good thread heng. i want to add to my fellow competitors that the literature has many variations on molecule representation , and we dont know the best. one paper i read said that graph theory/connected edge is one of the best representations. another says look into representing distances between atoms, and also angles between them. Also i saw a paper convert some of the \"molecule\" representations into \"atomic\" representations by instead of considering all atoms in the molecule, only consider neighbor atoms within a Radius of the atoms you are focusing on. The literature seems shotgunned with so many different molecule techniques that it will be exciting to see the variation of representations between the winners",
      "votes": null
    },
    {
      "id": "560684",
      "postDate": "06/25/2019 17:35:33",
      "content": "<p>ML learned descriptor:  </p>\n\n<p>End-to-end Symmetry Preserving Inter-atomic\nPotential Energy Model for Finite and Extended\nSystems\n<a href=\"https://arxiv.org/pdf/1805.09003.pdf\">https://arxiv.org/pdf/1805.09003.pdf</a></p>",
      "rawMarkdown": "ML learned descriptor:  \n\nEnd-to-end Symmetry Preserving Inter-atomic\nPotential Energy Model for Finite and Extended\nSystems\nhttps://arxiv.org/pdf/1805.09003.pdf",
      "votes": null
    },
    {
      "id": "566206",
      "postDate": "07/01/2019 22:31:26",
      "content": "<p>GRAPE as an graph extension to SOAP, </p>\n\n<p><a href=\"https://arxiv.org/pdf/1612.00193.pdf\">https://arxiv.org/pdf/1612.00193.pdf</a></p>",
      "rawMarkdown": "GRAPE as an graph extension to SOAP, \n\nhttps://arxiv.org/pdf/1612.00193.pdf",
      "votes": null
    },
    {
      "id": "571623",
      "postDate": "07/09/2019 21:07:59",
      "content": "<p>Are you using this to get molecule representations for the test data?</p>",
      "rawMarkdown": "Are you using this to get molecule representations for the test data?",
      "votes": null
    },
    {
      "id": "573748",
      "postDate": "07/12/2019 17:29:39",
      "content": "<p>I tried a Coulomb matrix but  it did not work well.\nI am trying other features implemented of dscribe. It takes time to use because of soap are many features.</p>",
      "rawMarkdown": "I tried a Coulomb matrix but  it did not work well.\nI am trying other features implemented of dscribe. It takes time to use because of soap are many features.",
      "votes": null
    },
    {
      "id": "573913",
      "postDate": "07/13/2019 00:15:13",
      "content": "<p>Coulomb matrix is very simple. Just two parameters, nuclear charge and distance are used. We should not expect high performance from it alone. </p>",
      "rawMarkdown": "Coulomb matrix is very simple. Just two parameters, nuclear charge and distance are used. We should not expect high performance from it alone.",
      "votes": null
    },
    {
      "id": "574289",
      "postDate": "07/13/2019 15:16:07",
      "content": "<p>Thank you!  :)</p>",
      "rawMarkdown": "Thank you!  :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 560045,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "06/25/2019 01:36:53",
      "content": "<p>Get angle information from rdkit\n<a href=\"https://www.google.com/amp/s/iwatobipen.wordpress.com/2016/07/16/get-molecular-angle-information-using-rdkit/amp/\">https://www.google.com/amp/s/iwatobipen.wordpress.com/2016/07/16/get-molecular-angle-information-using-rdkit/amp/</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 560107,
      "author_name": "returnofsputnik",
      "author_url": "",
      "post_date": "06/25/2019 03:56:26",
      "content": "<p>this is a very good thread heng. i want to add to my fellow competitors that the literature has many variations on molecule representation , and we dont know the best. one paper i read said that graph theory/connected edge is one of the best representations. another says look into representing distances between atoms, and also angles between them. Also i saw a paper convert some of the \"molecule\" representations into \"atomic\" representations by instead of considering all atoms in the molecule, only consider neighbor atoms within a Radius of the atoms you are focusing on. The literature seems shotgunned with so many different molecule techniques that it will be exciting to see the variation of representations between the winners</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 560684,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "06/25/2019 17:35:33",
      "content": "<p>ML learned descriptor:  </p>\n\n<p>End-to-end Symmetry Preserving Inter-atomic\nPotential Energy Model for Finite and Extended\nSystems\n<a href=\"https://arxiv.org/pdf/1805.09003.pdf\">https://arxiv.org/pdf/1805.09003.pdf</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 566206,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "07/01/2019 22:31:26",
      "content": "<p>GRAPE as an graph extension to SOAP, </p>\n\n<p><a href=\"https://arxiv.org/pdf/1612.00193.pdf\">https://arxiv.org/pdf/1612.00193.pdf</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 571623,
      "author_name": "skinish",
      "author_url": "",
      "post_date": "07/09/2019 21:07:59",
      "content": "<p>Are you using this to get molecule representations for the test data?</p>",
      "votes": null,
      "replies": [
        {
          "id": 573748,
          "author_name": "kani23",
          "author_url": "",
          "post_date": "07/12/2019 17:29:39",
          "content": "<p>I tried a Coulomb matrix but  it did not work well.\nI am trying other features implemented of dscribe. It takes time to use because of soap are many features.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 573913,
          "author_name": "matthewmasters",
          "author_url": "",
          "post_date": "07/13/2019 00:15:13",
          "content": "<p>Coulomb matrix is very simple. Just two parameters, nuclear charge and distance are used. We should not expect high performance from it alone. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 574289,
          "author_name": "kani23",
          "author_url": "",
          "post_date": "07/13/2019 15:16:07",
          "content": "<p>Thank you!  :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "559530": "https://github.com/SINGROUP/dscribe\n\n- Coulomb matrix\n- Sine matrix\n- Ewald matrix\n- Atom-centered Symmetry Functions (ACSF)\n- Smooth Overlap of Atomic Orbitals (SOAP)\n- Many-body Tensor Representation (MBTR)\n- Local Many-body Tensor Representation (LMBTR)\n\nhttps://github.com/atomistic-machine-learning/schnetpack\n - wACSF - weighted atom-centered symmetry functions",
    "560045": "Get angle information from rdkit\nhttps://www.google.com/amp/s/iwatobipen.wordpress.com/2016/07/16/get-molecular-angle-information-using-rdkit/amp/",
    "560107": "this is a very good thread heng. i want to add to my fellow competitors that the literature has many variations on molecule representation , and we dont know the best. one paper i read said that graph theory/connected edge is one of the best representations. another says look into representing distances between atoms, and also angles between them. Also i saw a paper convert some of the \"molecule\" representations into \"atomic\" representations by instead of considering all atoms in the molecule, only consider neighbor atoms within a Radius of the atoms you are focusing on. The literature seems shotgunned with so many different molecule techniques that it will be exciting to see the variation of representations between the winners",
    "560684": "ML learned descriptor:  \n\nEnd-to-end Symmetry Preserving Inter-atomic\nPotential Energy Model for Finite and Extended\nSystems\nhttps://arxiv.org/pdf/1805.09003.pdf",
    "566206": "GRAPE as an graph extension to SOAP, \n\nhttps://arxiv.org/pdf/1612.00193.pdf",
    "571623": "Are you using this to get molecule representations for the test data?",
    "573748": "I tried a Coulomb matrix but  it did not work well.\nI am trying other features implemented of dscribe. It takes time to use because of soap are many features.",
    "573913": "Coulomb matrix is very simple. Just two parameters, nuclear charge and distance are used. We should not expect high performance from it alone.",
    "574289": "Thank you!  :)"
  },
  "source": "meta"
}