{
  "id": 466703,
  "title": "Share: T5 embeddings for 500K+ proteins from the uniprot 2023-02 (reviewed only)",
  "url": "/competitions/cafa-5-protein-function-prediction/discussion/466703",
  "author_name": "",
  "post_date": "2024-01-09T17:11:01.150000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Kaggle dataset: <a href=\"https://www.kaggle.com/datasets/alexandervc/cafa5-t5-embeds-for-500k-uniprot-2023-02-reviewed\" target=\"_blank\">https://www.kaggle.com/datasets/alexandervc/cafa5-t5-embeds-for-500k-uniprot-2023-02-reviewed</a><br>\nT5 embeddings for 500K+ proteins from the uniprot 2023-02 (reviewed only)</p>\n<p>Embedings for more than 500K proteins from the uniprot 2023-02 (reviewed only - see Dataset <a href=\"https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only\" target=\"_blank\">https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only</a><br>\nand particular fasta file: <a href=\"https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.fa\" target=\"_blank\">https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.fa</a><br>\nCorresponding GO terms in the file: <a href=\"https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.DR.GO.table\" target=\"_blank\">https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.DR.GO.table</a> )</p>\n<p>Embeddings computed by Sergei Fironov shared with his permission.</p>\n<hr>\n<p>Our team tried to use these embeds   - pretrain the model on that large dataset - got some CV uplift, but fail on LB, so not used in the final submits. </p>\n<hr>\n<p>PS</p>\n<p>There is a month newer version of uniprot sequences: <br>\n<a href=\"https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-03-reviewed-only\" target=\"_blank\">https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-03-reviewed-only</a><br>\nbut we have not computed embeds for it. Any way difference is not big.</p>",
  "messages": [
    {
      "id": 2594130,
      "postDate": "2024-01-09T17:11:01.150Z",
      "content": "<p>Kaggle dataset: <a href=\"https://www.kaggle.com/datasets/alexandervc/cafa5-t5-embeds-for-500k-uniprot-2023-02-reviewed\" target=\"_blank\">https://www.kaggle.com/datasets/alexandervc/cafa5-t5-embeds-for-500k-uniprot-2023-02-reviewed</a><br>\nT5 embeddings for 500K+ proteins from the uniprot 2023-02 (reviewed only)</p>\n<p>Embedings for more than 500K proteins from the uniprot 2023-02 (reviewed only - see Dataset <a href=\"https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only\" target=\"_blank\">https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only</a><br>\nand particular fasta file: <a href=\"https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.fa\" target=\"_blank\">https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.fa</a><br>\nCorresponding GO terms in the file: <a href=\"https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.DR.GO.table\" target=\"_blank\">https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.DR.GO.table</a> )</p>\n<p>Embeddings computed by Sergei Fironov shared with his permission.</p>\n<hr>\n<p>Our team tried to use these embeds   - pretrain the model on that large dataset - got some CV uplift, but fail on LB, so not used in the final submits. </p>\n<hr>\n<p>PS</p>\n<p>There is a month newer version of uniprot sequences: <br>\n<a href=\"https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-03-reviewed-only\" target=\"_blank\">https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-03-reviewed-only</a><br>\nbut we have not computed embeds for it. Any way difference is not big.</p>",
      "rawMarkdown": "Kaggle dataset: https://www.kaggle.com/datasets/alexandervc/cafa5-t5-embeds-for-500k-uniprot-2023-02-reviewed\nT5 embeddings for 500K+ proteins from the uniprot 2023-02 (reviewed only)\n\nEmbedings for more than 500K proteins from the uniprot 2023-02 (reviewed only - see Dataset https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only\nand particular fasta file: https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.fa\nCorresponding GO terms in the file: https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.DR.GO.table )\n\n\nEmbeddings computed by Sergei Fironov shared with his permission.\n\n-------------\n\nOur team tried to use these embeds   - pretrain the model on that large dataset - got some CV uplift, but fail on LB, so not used in the final submits. \n\n-------------\n\nPS\n\nThere is a month newer version of uniprot sequences: \nhttps://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-03-reviewed-only\nbut we have not computed embeds for it. Any way difference is not big.\n",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2594130": "Kaggle dataset: https://www.kaggle.com/datasets/alexandervc/cafa5-t5-embeds-for-500k-uniprot-2023-02-reviewed\nT5 embeddings for 500K+ proteins from the uniprot 2023-02 (reviewed only)\n\nEmbedings for more than 500K proteins from the uniprot 2023-02 (reviewed only - see Dataset https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only\nand particular fasta file: https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.fa\nCorresponding GO terms in the file: https://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-02-reviewed-only/data?select=uniprot.DR.GO.table )\n\n\nEmbeddings computed by Sergei Fironov shared with his permission.\n\n-------------\n\nOur team tried to use these embeds   - pretrain the model on that large dataset - got some CV uplift, but fail on LB, so not used in the final submits. \n\n-------------\n\nPS\n\nThere is a month newer version of uniprot sequences: \nhttps://www.kaggle.com/datasets/yevgeniyantipin/uniprot-2023-03-reviewed-only\nbut we have not computed embeds for it. Any way difference is not big.\n"
  }
}