{
  "id": 349162,
  "title": "Why Multiome part is so hard than CITEseq part? ",
  "url": "/competitions/open-problems-multimodal/discussion/349162",
  "author_name": "",
  "post_date": "2022-08-31T12:49:12.475101700Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I know you think I talk about the difficulty of using huge data from the Multiome part, but even if we have huge numbers of features and a higher size of rows, I found the best 500 features in the Multiome part have an absolute correlation between 5% to 1% with targets.<br>\nIn CITEseq part, we have the best 500 features that have an absolute correlation between 47% to 6% with targets.<br>\nI don't know why we have a huge difference between the two parts, </p>",
  "messages": [
    {
      "id": "1920838",
      "postDate": "08/31/2022 12:49:12",
      "content": "<p>I know you think I talk about the difficulty of using huge data from the Multiome part, but even if we have huge numbers of features and a higher size of rows, I found the best 500 features in the Multiome part have an absolute correlation between 5% to 1% with targets.<br>\nIn CITEseq part, we have the best 500 features that have an absolute correlation between 47% to 6% with targets.<br>\nI don't know why we have a huge difference between the two parts, </p>",
      "rawMarkdown": "I know you think I talk about the difficulty of using huge data from the Multiome part, but even if we have huge numbers of features and a higher size of rows, I found the best 500 features in the Multiome part have an absolute correlation between 5% to 1% with targets.\nIn CITEseq part, we have the best 500 features that have an absolute correlation between 47% to 6% with targets.\nI don't know why we have a huge difference between the two parts,",
      "votes": null
    },
    {
      "id": "1921170",
      "postDate": "08/31/2022 16:14:45",
      "content": "<p>Booba, how do you derive your 500 features here?</p>",
      "rawMarkdown": "Booba, how do you derive your 500 features here?",
      "votes": null
    },
    {
      "id": "1921197",
      "postDate": "08/31/2022 16:26:32",
      "content": "<p>By Absolute Correlation, like this <br>\n<code>abs(np.corrcoef( df_cite_train_y[features_y],  df_cite_train_x[features_x])[0,1])</code><br>\nYou can do a loop to summarize all Absolute Correlations for each target.</p>",
      "rawMarkdown": "By Absolute Correlation, like this \n`abs(np.corrcoef( df_cite_train_y[features_y],  df_cite_train_x[features_x])[0,1])`\nYou can do a loop to summarize all Absolute Correlations for each target.",
      "votes": null
    },
    {
      "id": "1921743",
      "postDate": "09/01/2022 03:04:22",
      "content": "<p>I found multiome target genes contain lots of biologically unknown genes and long noncoding RNAs. Could that be the reason?<br>\n<a href=\"https://www.kaggle.com/code/masato114/overlaps-between-cite-genes-and-atac-genes/notebook\" target=\"_blank\">Notebook</a> was shared right now. I am glad if you find this helpful.</p>",
      "rawMarkdown": "I found multiome target genes contain lots of biologically unknown genes and long noncoding RNAs. Could that be the reason?\n[Notebook](https://www.kaggle.com/code/masato114/overlaps-between-cite-genes-and-atac-genes/notebook) was shared right now. I am glad if you find this helpful.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1921170,
      "author_name": "drpatrickchan",
      "author_url": "",
      "post_date": "08/31/2022 16:14:45",
      "content": "<p>Booba, how do you derive your 500 features here?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1921197,
          "author_name": "youneseloiarm",
          "author_url": "",
          "post_date": "08/31/2022 16:26:32",
          "content": "<p>By Absolute Correlation, like this <br>\n<code>abs(np.corrcoef( df_cite_train_y[features_y],  df_cite_train_x[features_x])[0,1])</code><br>\nYou can do a loop to summarize all Absolute Correlations for each target.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1921743,
      "author_name": "masato114",
      "author_url": "",
      "post_date": "09/01/2022 03:04:22",
      "content": "<p>I found multiome target genes contain lots of biologically unknown genes and long noncoding RNAs. Could that be the reason?<br>\n<a href=\"https://www.kaggle.com/code/masato114/overlaps-between-cite-genes-and-atac-genes/notebook\" target=\"_blank\">Notebook</a> was shared right now. I am glad if you find this helpful.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1920838": "I know you think I talk about the difficulty of using huge data from the Multiome part, but even if we have huge numbers of features and a higher size of rows, I found the best 500 features in the Multiome part have an absolute correlation between 5% to 1% with targets.\nIn CITEseq part, we have the best 500 features that have an absolute correlation between 47% to 6% with targets.\nI don't know why we have a huge difference between the two parts,",
    "1921170": "Booba, how do you derive your 500 features here?",
    "1921197": "By Absolute Correlation, like this \n`abs(np.corrcoef( df_cite_train_y[features_y],  df_cite_train_x[features_x])[0,1])`\nYou can do a loop to summarize all Absolute Correlations for each target.",
    "1921743": "I found multiome target genes contain lots of biologically unknown genes and long noncoding RNAs. Could that be the reason?\n[Notebook](https://www.kaggle.com/code/masato114/overlaps-between-cite-genes-and-atac-genes/notebook) was shared right now. I am glad if you find this helpful."
  },
  "source": "meta"
}