{
  "id": 568123,
  "title": "RibonanzaNet Explained",
  "url": "/competitions/stanford-rna-3d-folding/discussion/568123",
  "author_name": "",
  "post_date": "2025-03-13T23:37:32.675857400Z",
  "votes": 16,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I made a tutorial where I explain all the building blocks that compose the <code>RibonanzaNet</code> architecture, proposed by <a href=\"https://www.kaggle.com/shujun717\" target=\"_blank\">@shujun717</a> (our competition's host) et al. This architecture unifies features of <code>RNAdegformer</code> and top Kaggle models (from last year's Stanford Ribonanza RNA Folding competition) into a single, self-contained model.</p>\n<p>You will find all the building blocks involved, their descriptions, purposes and diagrams. Each class is detailed with the different input definitions, tensor shapes and data types to better understand them.</p>\n<p>You can find the tutorial here:</p>\n<ul>\n<li><a href=\"https://www.kaggle.com/code/alejopaullier/stanford-rna-3d-ribonanzanet-explained\" target=\"_blank\">Stanford RNA 3D | RibonanzaNet Explained</a></li>\n</ul>\n<p>Please leave a comment if you have any suggestions.</p>\n<p>Best of luck in the competition! 🍀 </p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3197853%2F91f2210e245272387e4744a89d13839b%2Fribonanza_diagram.png?generation=1741908727212758&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "3149202",
      "postDate": "03/13/2025 23:37:32",
      "content": "<p>I made a tutorial where I explain all the building blocks that compose the <code>RibonanzaNet</code> architecture, proposed by <a href=\"https://www.kaggle.com/shujun717\" target=\"_blank\">@shujun717</a> (our competition's host) et al. This architecture unifies features of <code>RNAdegformer</code> and top Kaggle models (from last year's Stanford Ribonanza RNA Folding competition) into a single, self-contained model.</p>\n<p>You will find all the building blocks involved, their descriptions, purposes and diagrams. Each class is detailed with the different input definitions, tensor shapes and data types to better understand them.</p>\n<p>You can find the tutorial here:</p>\n<ul>\n<li><a href=\"https://www.kaggle.com/code/alejopaullier/stanford-rna-3d-ribonanzanet-explained\" target=\"_blank\">Stanford RNA 3D | RibonanzaNet Explained</a></li>\n</ul>\n<p>Please leave a comment if you have any suggestions.</p>\n<p>Best of luck in the competition! 🍀 </p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3197853%2F91f2210e245272387e4744a89d13839b%2Fribonanza_diagram.png?generation=1741908727212758&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "I made a tutorial where I explain all the building blocks that compose the `RibonanzaNet` architecture, proposed by @shujun717 (our competition's host) et al. This architecture unifies features of `RNAdegformer` and top Kaggle models (from last year's Stanford Ribonanza RNA Folding competition) into a single, self-contained model.\n\nYou will find all the building blocks involved, their descriptions, purposes and diagrams. Each class is detailed with the different input definitions, tensor shapes and data types to better understand them.\n\nYou can find the tutorial here:\n- [Stanford RNA 3D | RibonanzaNet Explained](https://www.kaggle.com/code/alejopaullier/stanford-rna-3d-ribonanzanet-explained)\n\nPlease leave a comment if you have any suggestions.\n\nBest of luck in the competition! 🍀 \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3197853%2F91f2210e245272387e4744a89d13839b%2Fribonanza_diagram.png?generation=1741908727212758&alt=media)",
      "votes": null
    },
    {
      "id": "3149467",
      "postDate": "03/14/2025 08:57:29",
      "content": "<p><a href=\"https://www.youtube.com/watch?v=XqFq_zYx7Vo&amp;t=4136s\" target=\"_blank\">https://www.youtube.com/watch?v=XqFq_zYx7Vo&amp;t=4136s</a><br>\nRhiju Das - Ribonanza Net <br>\n(i highly recommend to google and watch other rna videos by Prof Das as well)</p>\n<p><a href=\"https://www.youtube.com/watch?v=a8a41m_iCek&amp;t=3454s\" target=\"_blank\">https://www.youtube.com/watch?v=a8a41m_iCek&amp;t=3454s</a><br>\n<a href=\"https://www.youtube.com/watch?v=XXZIznnHMrM&amp;t=1597s\" target=\"_blank\">https://www.youtube.com/watch?v=XXZIznnHMrM&amp;t=1597s</a><br>\nShujun He - Ribonanza Net</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2F1dee839902a1fcf6d86091c146c0f790%2FSelection_050.png?generation=1741942579939079&amp;alt=media\" alt=\"\"></p>\n<p>Ribonanza Net is close to alphafold2 evoformer, so these maybe helpful<br>\n<a href=\"https://www.youtube.com/watch?v=gY4-vVRTkpk&amp;t=261s\" target=\"_blank\">https://www.youtube.com/watch?v=gY4-vVRTkpk&amp;t=261s</a><br>\nAlphaFold Decoded: Evoformer (Lesson 5)</p>\n<p><a href=\"https://www.youtube.com/watch?v=CYncNBMPLLk\" target=\"_blank\">https://www.youtube.com/watch?v=CYncNBMPLLk</a><br>\nThe brilliance of AlphaFold 3</p>",
      "rawMarkdown": "https://www.youtube.com/watch?v=XqFq_zYx7Vo&t=4136s\nRhiju Das - Ribonanza Net \n(i highly recommend to google and watch other rna videos by Prof Das as well)\n\nhttps://www.youtube.com/watch?v=a8a41m_iCek&t=3454s\nhttps://www.youtube.com/watch?v=XXZIznnHMrM&t=1597s\nShujun He - Ribonanza Net\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2F1dee839902a1fcf6d86091c146c0f790%2FSelection_050.png?generation=1741942579939079&alt=media)\n\n\nRibonanza Net is close to alphafold2 evoformer, so these maybe helpful\nhttps://www.youtube.com/watch?v=gY4-vVRTkpk&t=261s\nAlphaFold Decoded: Evoformer (Lesson 5)\n\nhttps://www.youtube.com/watch?v=CYncNBMPLLk\nThe brilliance of AlphaFold 3",
      "votes": null
    },
    {
      "id": "3149621",
      "postDate": "03/14/2025 13:30:23",
      "content": "<p>Thanks for the videos <a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a> Yes, the <code>ConvTransformerEncoderLayer</code> from <code>RibonanzaNet</code> indeed looks pretty similar to the <code>Evoformer</code> block from AlphaFold. In the notebook I stated the similarities between both. The difference lies mostly on the initial attention mechanism (multihead vs row-wise and column-wise attention).</p>\n<p>I will check the videos!</p>",
      "rawMarkdown": "Thanks for the videos @hengck23 Yes, the `ConvTransformerEncoderLayer` from `RibonanzaNet` indeed looks pretty similar to the `Evoformer` block from AlphaFold. In the notebook I stated the similarities between both. The difference lies mostly on the initial attention mechanism (multihead vs row-wise and column-wise attention).\n\nI will check the videos!",
      "votes": null
    },
    {
      "id": "3155667",
      "postDate": "03/21/2025 09:27:04",
      "content": "<p><a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a> thanks for sharing! It's very interesting</p>",
      "rawMarkdown": "hengck23 thanks for sharing! It's very interesting",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3149467,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "03/14/2025 08:57:29",
      "content": "<p><a href=\"https://www.youtube.com/watch?v=XqFq_zYx7Vo&amp;t=4136s\" target=\"_blank\">https://www.youtube.com/watch?v=XqFq_zYx7Vo&amp;t=4136s</a><br>\nRhiju Das - Ribonanza Net <br>\n(i highly recommend to google and watch other rna videos by Prof Das as well)</p>\n<p><a href=\"https://www.youtube.com/watch?v=a8a41m_iCek&amp;t=3454s\" target=\"_blank\">https://www.youtube.com/watch?v=a8a41m_iCek&amp;t=3454s</a><br>\n<a href=\"https://www.youtube.com/watch?v=XXZIznnHMrM&amp;t=1597s\" target=\"_blank\">https://www.youtube.com/watch?v=XXZIznnHMrM&amp;t=1597s</a><br>\nShujun He - Ribonanza Net</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2F1dee839902a1fcf6d86091c146c0f790%2FSelection_050.png?generation=1741942579939079&amp;alt=media\" alt=\"\"></p>\n<p>Ribonanza Net is close to alphafold2 evoformer, so these maybe helpful<br>\n<a href=\"https://www.youtube.com/watch?v=gY4-vVRTkpk&amp;t=261s\" target=\"_blank\">https://www.youtube.com/watch?v=gY4-vVRTkpk&amp;t=261s</a><br>\nAlphaFold Decoded: Evoformer (Lesson 5)</p>\n<p><a href=\"https://www.youtube.com/watch?v=CYncNBMPLLk\" target=\"_blank\">https://www.youtube.com/watch?v=CYncNBMPLLk</a><br>\nThe brilliance of AlphaFold 3</p>",
      "votes": null,
      "replies": [
        {
          "id": 3149621,
          "author_name": "alejopaullier",
          "author_url": "",
          "post_date": "03/14/2025 13:30:23",
          "content": "<p>Thanks for the videos <a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a> Yes, the <code>ConvTransformerEncoderLayer</code> from <code>RibonanzaNet</code> indeed looks pretty similar to the <code>Evoformer</code> block from AlphaFold. In the notebook I stated the similarities between both. The difference lies mostly on the initial attention mechanism (multihead vs row-wise and column-wise attention).</p>\n<p>I will check the videos!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 3155667,
          "author_name": "volodymyrpivoshenko",
          "author_url": "",
          "post_date": "03/21/2025 09:27:04",
          "content": "<p><a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a> thanks for sharing! It's very interesting</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3149202": "I made a tutorial where I explain all the building blocks that compose the `RibonanzaNet` architecture, proposed by @shujun717 (our competition's host) et al. This architecture unifies features of `RNAdegformer` and top Kaggle models (from last year's Stanford Ribonanza RNA Folding competition) into a single, self-contained model.\n\nYou will find all the building blocks involved, their descriptions, purposes and diagrams. Each class is detailed with the different input definitions, tensor shapes and data types to better understand them.\n\nYou can find the tutorial here:\n- [Stanford RNA 3D | RibonanzaNet Explained](https://www.kaggle.com/code/alejopaullier/stanford-rna-3d-ribonanzanet-explained)\n\nPlease leave a comment if you have any suggestions.\n\nBest of luck in the competition! 🍀 \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3197853%2F91f2210e245272387e4744a89d13839b%2Fribonanza_diagram.png?generation=1741908727212758&alt=media)",
    "3149467": "https://www.youtube.com/watch?v=XqFq_zYx7Vo&t=4136s\nRhiju Das - Ribonanza Net \n(i highly recommend to google and watch other rna videos by Prof Das as well)\n\nhttps://www.youtube.com/watch?v=a8a41m_iCek&t=3454s\nhttps://www.youtube.com/watch?v=XXZIznnHMrM&t=1597s\nShujun He - Ribonanza Net\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2F1dee839902a1fcf6d86091c146c0f790%2FSelection_050.png?generation=1741942579939079&alt=media)\n\n\nRibonanza Net is close to alphafold2 evoformer, so these maybe helpful\nhttps://www.youtube.com/watch?v=gY4-vVRTkpk&t=261s\nAlphaFold Decoded: Evoformer (Lesson 5)\n\nhttps://www.youtube.com/watch?v=CYncNBMPLLk\nThe brilliance of AlphaFold 3",
    "3149621": "Thanks for the videos @hengck23 Yes, the `ConvTransformerEncoderLayer` from `RibonanzaNet` indeed looks pretty similar to the `Evoformer` block from AlphaFold. In the notebook I stated the similarities between both. The difference lies mostly on the initial attention mechanism (multihead vs row-wise and column-wise attention).\n\nI will check the videos!",
    "3155667": "hengck23 thanks for sharing! It's very interesting"
  },
  "source": "meta"
}