{
  "id": 101950,
  "title": "Problem structures",
  "url": "/competitions/champs-scalar-coupling/discussion/101950",
  "author_name": "Lt. Cmdr. Data",
  "post_date": "2019-07-30T02:52:34.218000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Just thought I would point out there are at least 341 \"problem structures\" that rdkit can't sanitize, so a little over 0.2% of the dataset; missing hydrogens, weird valence and other issues. It's not an issue with file conversion, because if you load the original .xyz file in (insert favorite molecular viewer here), it's pretty obvious something went wrong. Probably there was an issue during geometry optimization for some of these. It's pretty typical IME when dealing with large bio/cheminformatics datasets to have a few bad eggs, but good to be aware of. Here are a few examples:</p>\n\n<p>dsgdb_9nsd025859, missing hydrogens / weird valence\ndsgdb_9nsd054885, missing hydrogens / weird valence\ndsgdb_9nsd_081568, something weird went wrong here...\ndsgdb9nsd_129579, missing h's again</p>\n\n<p>if I can find a simple way to fix these problem children without screwing up the atom numbering (sort of important, because that's how the datapoints are tracked to specific H and C), I'll upload the fixed xyz files here. Just thought it would be good for anyone who is just using the raw xyz files to generate features to be aware these are out there...</p>",
  "messages": [
    {
      "id": 588009,
      "postDate": "2019-07-30T02:52:34.217Z",
      "content": "<p>Just thought I would point out there are at least 341 \"problem structures\" that rdkit can't sanitize, so a little over 0.2% of the dataset; missing hydrogens, weird valence and other issues. It's not an issue with file conversion, because if you load the original .xyz file in (insert favorite molecular viewer here), it's pretty obvious something went wrong. Probably there was an issue during geometry optimization for some of these. It's pretty typical IME when dealing with large bio/cheminformatics datasets to have a few bad eggs, but good to be aware of. Here are a few examples:</p>\n\n<p>dsgdb_9nsd025859, missing hydrogens / weird valence\ndsgdb_9nsd054885, missing hydrogens / weird valence\ndsgdb_9nsd_081568, something weird went wrong here...\ndsgdb9nsd_129579, missing h's again</p>\n\n<p>if I can find a simple way to fix these problem children without screwing up the atom numbering (sort of important, because that's how the datapoints are tracked to specific H and C), I'll upload the fixed xyz files here. Just thought it would be good for anyone who is just using the raw xyz files to generate features to be aware these are out there...</p>",
      "rawMarkdown": "Just thought I would point out there are at least 341 \"problem structures\" that rdkit can't sanitize, so a little over 0.2% of the dataset; missing hydrogens, weird valence and other issues. It's not an issue with file conversion, because if you load the original .xyz file in (insert favorite molecular viewer here), it's pretty obvious something went wrong. Probably there was an issue during geometry optimization for some of these. It's pretty typical IME when dealing with large bio/cheminformatics datasets to have a few bad eggs, but good to be aware of. Here are a few examples:\n\ndsgdb_9nsd025859, missing hydrogens / weird valence\ndsgdb_9nsd054885, missing hydrogens / weird valence\ndsgdb_9nsd_081568, something weird went wrong here...\ndsgdb9nsd_129579, missing h's again\n\nif I can find a simple way to fix these problem children without screwing up the atom numbering (sort of important, because that's how the datapoints are tracked to specific H and C), I'll upload the fixed xyz files here. Just thought it would be good for anyone who is just using the raw xyz files to generate features to be aware these are out there...",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "588009": "Just thought I would point out there are at least 341 \"problem structures\" that rdkit can't sanitize, so a little over 0.2% of the dataset; missing hydrogens, weird valence and other issues. It's not an issue with file conversion, because if you load the original .xyz file in (insert favorite molecular viewer here), it's pretty obvious something went wrong. Probably there was an issue during geometry optimization for some of these. It's pretty typical IME when dealing with large bio/cheminformatics datasets to have a few bad eggs, but good to be aware of. Here are a few examples:\n\ndsgdb_9nsd025859, missing hydrogens / weird valence\ndsgdb_9nsd054885, missing hydrogens / weird valence\ndsgdb_9nsd_081568, something weird went wrong here...\ndsgdb9nsd_129579, missing h's again\n\nif I can find a simple way to fix these problem children without screwing up the atom numbering (sort of important, because that's how the datapoints are tracked to specific H and C), I'll upload the fixed xyz files here. Just thought it would be good for anyone who is just using the raw xyz files to generate features to be aware these are out there..."
  }
}