{
  "id": 456874,
  "title": "Question Regarding DE model",
  "url": "/competitions/open-problems-single-cell-perturbations/discussion/456874",
  "author_name": "Kishan Vavdara",
  "post_date": "2023-11-22T06:36:48.116000",
  "votes": 11,
  "comment_count": 0,
  "views": 0,
  "content": "<p>The description mentions that for training, the DE model is fit once to all samples from the training set of (compound, cell type) combinations. However, for both the public and private tests, the DE model is fit to all data.</p>\n<p>I wonder how training the DE model on the full data, as opposed to using only train set (compound, cell type) combinations during training, might impact the DE for genes. Could this potentially introduce variations in the&nbsp;DE&nbsp;for&nbsp;genes?</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F12951282%2F30a4f323c51906f0dadda6d81c257bbe%2FScreenshot%202023-11-22%20120334.png?generation=1700634914296378&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": 2533681,
      "postDate": "2023-11-22T06:36:48.117Z",
      "content": "<p>The description mentions that for training, the DE model is fit once to all samples from the training set of (compound, cell type) combinations. However, for both the public and private tests, the DE model is fit to all data.</p>\n<p>I wonder how training the DE model on the full data, as opposed to using only train set (compound, cell type) combinations during training, might impact the DE for genes. Could this potentially introduce variations in the&nbsp;DE&nbsp;for&nbsp;genes?</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F12951282%2F30a4f323c51906f0dadda6d81c257bbe%2FScreenshot%202023-11-22%20120334.png?generation=1700634914296378&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "The description mentions that for training, the DE model is fit once to all samples from the training set of (compound, cell type) combinations. However, for both the public and private tests, the DE model is fit to all data.\n\nI wonder how training the DE model on the full data, as opposed to using only train set (compound, cell type) combinations during training, might impact the DE for genes. Could this potentially introduce variations in the DE for genes?\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F12951282%2F30a4f323c51906f0dadda6d81c257bbe%2FScreenshot%202023-11-22%20120334.png?generation=1700634914296378&alt=media)",
      "votes": 11
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2533681": "The description mentions that for training, the DE model is fit once to all samples from the training set of (compound, cell type) combinations. However, for both the public and private tests, the DE model is fit to all data.\n\nI wonder how training the DE model on the full data, as opposed to using only train set (compound, cell type) combinations during training, might impact the DE for genes. Could this potentially introduce variations in the DE for genes?\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F12951282%2F30a4f323c51906f0dadda6d81c257bbe%2FScreenshot%202023-11-22%20120334.png?generation=1700634914296378&alt=media)"
  }
}