{
  "id": 664979,
  "title": "What is a useful toy problem to speed up prototyping?",
  "url": "/competitions/cafa-6-protein-function-prediction/discussion/664979",
  "author_name": "",
  "post_date": "2025-12-29T11:13:42.792568500Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I'm not even sure if it makes sense to ask this question.</p>\n<p>But trying to replicate the full problem with all the GO labels and hierarchy makes it hard to 'try out' things quickly. </p>\n<p>I'm wondering if there are good toy problems to speed this up that still captures the qualitative nature of the problem (hierarchy etc.)</p>\n<p>For example, maybe just predicting the GO label root nodes (and one level lower). Incorporating key words? Does a term belong to some subgraph? </p>",
  "messages": [
    {
      "id": "3383085",
      "postDate": "12/29/2025 11:13:42",
      "content": "<p>I'm not even sure if it makes sense to ask this question.</p>\n<p>But trying to replicate the full problem with all the GO labels and hierarchy makes it hard to 'try out' things quickly. </p>\n<p>I'm wondering if there are good toy problems to speed this up that still captures the qualitative nature of the problem (hierarchy etc.)</p>\n<p>For example, maybe just predicting the GO label root nodes (and one level lower). Incorporating key words? Does a term belong to some subgraph? </p>",
      "rawMarkdown": "I'm not even sure if it makes sense to ask this question.\n\nBut trying to replicate the full problem with all the GO labels and hierarchy makes it hard to 'try out' things quickly. \n\nI'm wondering if there are good toy problems to speed this up that still captures the qualitative nature of the problem (hierarchy etc.)\n\nFor example, maybe just predicting the GO label root nodes (and one level lower). Incorporating key words? Does a term belong to some subgraph?",
      "votes": null
    },
    {
      "id": "3383224",
      "postDate": "12/29/2025 16:48:37",
      "content": "<p>I'm leaning towards just focusing on the C part of the output since it is significantly smaller than the other two. </p>",
      "rawMarkdown": "I'm leaning towards just focusing on the C part of the output since it is significantly smaller than the other two.",
      "votes": null
    },
    {
      "id": "3385694",
      "postDate": "01/03/2026 17:59:24",
      "content": "<p>A tool that I’ve found genuinely useful for rapid qualitative feedback is <strong>QuickGO</strong> (<a href=\"https://www.ebi.ac.uk/QuickGO/\" target=\"_blank\">https://www.ebi.ac.uk/QuickGO/</a>).</p>\n<p>Beyond whatever evaluation I’m already doing (via submissions or local metrics), my typical iteration workflow is:</p>\n<ul>\n<li>Take one of a few “known” proteins whose annotation topology I roughly understand.</li>\n<li>Look it up in QuickGO.</li>\n<li>Add to the basket the GO terms predicted by my method (or selected manually / semi-automatically).</li>\n<li>Visually inspect which terms are correct, which are wrong, and—crucially—how far the wrong ones are in the ontology.</li>\n</ul>\n<p>Doing this by GO category and by exploring different branches keeps manual inspection tractable, while still preserving what matters qualitatively: hierarchy, proximity in the DAG, and over- vs under-specific predictions.</p>\n<p>But of course, there are so many cases that, after months of inspecting proteins and their annotations, I’ve found myself unable to distill any clear, general rule for pruning annotations in a principled way. Still, without this kind of exercise, it’s very hard to build any real intuition about how the ontology actually behaves.</p>",
      "rawMarkdown": "A tool that I’ve found genuinely useful for rapid qualitative feedback is **QuickGO** ([https://www.ebi.ac.uk/QuickGO/](https://www.ebi.ac.uk/QuickGO/)).\n\nBeyond whatever evaluation I’m already doing (via submissions or local metrics), my typical iteration workflow is:\n\n* Take one of a few “known” proteins whose annotation topology I roughly understand.\n* Look it up in QuickGO.\n* Add to the basket the GO terms predicted by my method (or selected manually / semi-automatically).\n* Visually inspect which terms are correct, which are wrong, and—crucially—how far the wrong ones are in the ontology.\n\nDoing this by GO category and by exploring different branches keeps manual inspection tractable, while still preserving what matters qualitatively: hierarchy, proximity in the DAG, and over- vs under-specific predictions.\n\nBut of course, there are so many cases that, after months of inspecting proteins and their annotations, I’ve found myself unable to distill any clear, general rule for pruning annotations in a principled way. Still, without this kind of exercise, it’s very hard to build any real intuition about how the ontology actually behaves.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3383224,
      "author_name": "glennkroegel",
      "author_url": "",
      "post_date": "12/29/2025 16:48:37",
      "content": "<p>I'm leaning towards just focusing on the C part of the output since it is significantly smaller than the other two. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3385694,
      "author_name": "xaxipirulazo",
      "author_url": "",
      "post_date": "01/03/2026 17:59:24",
      "content": "<p>A tool that I’ve found genuinely useful for rapid qualitative feedback is <strong>QuickGO</strong> (<a href=\"https://www.ebi.ac.uk/QuickGO/\" target=\"_blank\">https://www.ebi.ac.uk/QuickGO/</a>).</p>\n<p>Beyond whatever evaluation I’m already doing (via submissions or local metrics), my typical iteration workflow is:</p>\n<ul>\n<li>Take one of a few “known” proteins whose annotation topology I roughly understand.</li>\n<li>Look it up in QuickGO.</li>\n<li>Add to the basket the GO terms predicted by my method (or selected manually / semi-automatically).</li>\n<li>Visually inspect which terms are correct, which are wrong, and—crucially—how far the wrong ones are in the ontology.</li>\n</ul>\n<p>Doing this by GO category and by exploring different branches keeps manual inspection tractable, while still preserving what matters qualitatively: hierarchy, proximity in the DAG, and over- vs under-specific predictions.</p>\n<p>But of course, there are so many cases that, after months of inspecting proteins and their annotations, I’ve found myself unable to distill any clear, general rule for pruning annotations in a principled way. Still, without this kind of exercise, it’s very hard to build any real intuition about how the ontology actually behaves.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3383085": "I'm not even sure if it makes sense to ask this question.\n\nBut trying to replicate the full problem with all the GO labels and hierarchy makes it hard to 'try out' things quickly. \n\nI'm wondering if there are good toy problems to speed this up that still captures the qualitative nature of the problem (hierarchy etc.)\n\nFor example, maybe just predicting the GO label root nodes (and one level lower). Incorporating key words? Does a term belong to some subgraph?",
    "3383224": "I'm leaning towards just focusing on the C part of the output since it is significantly smaller than the other two.",
    "3385694": "A tool that I’ve found genuinely useful for rapid qualitative feedback is **QuickGO** ([https://www.ebi.ac.uk/QuickGO/](https://www.ebi.ac.uk/QuickGO/)).\n\nBeyond whatever evaluation I’m already doing (via submissions or local metrics), my typical iteration workflow is:\n\n* Take one of a few “known” proteins whose annotation topology I roughly understand.\n* Look it up in QuickGO.\n* Add to the basket the GO terms predicted by my method (or selected manually / semi-automatically).\n* Visually inspect which terms are correct, which are wrong, and—crucially—how far the wrong ones are in the ontology.\n\nDoing this by GO category and by exploring different branches keeps manual inspection tractable, while still preserving what matters qualitatively: hierarchy, proximity in the DAG, and over- vs under-specific predictions.\n\nBut of course, there are so many cases that, after months of inspecting proteins and their annotations, I’ve found myself unable to distill any clear, general rule for pruning annotations in a principled way. Still, without this kind of exercise, it’s very hard to build any real intuition about how the ontology actually behaves."
  },
  "source": "meta"
}