Hi Kagglers,
\nWe’re happy to announce the conclusion of the CAFA 6 Protein Function Prediction competition. Thank you everyone for your participation!
\nThis competition ended with 9,249 registrations and 2,538 participants on 2,172 teams. We had 3,747 submissions from 103 countries. For 631 users (including 31 in the top 100!), this was their first competition. Thank you all for your hard work in this competition and congratulations to our winners and to those who gained a new ranking!
\nThe top potential winning teams will be contacted via email for the next steps. We look forward to learning more about their winning solutions.
\nWe've cleaned the leaderboard and disqualified some teams that have violated the rules. If you think you were removed by mistake, or believe you have evidence that suggests another team cheated, please contact compliance. Please fill in all the fields honestly.
\nWe highly encourage you to post a solution write-up about your approach and solution in the forums (see instructions). You may also refer to Kaggle Solution Write-Up Documentation for guidance. You are also encouraged to publish your models on Kaggle Models!
\nThanks for continuing to make Kaggle a great place to learn, practice, and test data science techniques!
\nHappy Modeling!
\nKaggle Team
", "messages": [ { "id": "3471426", "postDate": "06/11/2026 14:41:27", "content": "Hi Kagglers,
\nWe’re happy to announce the conclusion of the CAFA 6 Protein Function Prediction competition. Thank you everyone for your participation!
\nThis competition ended with 9,249 registrations and 2,538 participants on 2,172 teams. We had 3,747 submissions from 103 countries. For 631 users (including 31 in the top 100!), this was their first competition. Thank you all for your hard work in this competition and congratulations to our winners and to those who gained a new ranking!
\nThe top potential winning teams will be contacted via email for the next steps. We look forward to learning more about their winning solutions.
\nWe've cleaned the leaderboard and disqualified some teams that have violated the rules. If you think you were removed by mistake, or believe you have evidence that suggests another team cheated, please contact compliance. Please fill in all the fields honestly.
\nWe highly encourage you to post a solution write-up about your approach and solution in the forums (see instructions). You may also refer to Kaggle Solution Write-Up Documentation for guidance. You are also encouraged to publish your models on Kaggle Models!
\nThanks for continuing to make Kaggle a great place to learn, practice, and test data science techniques!
\nHappy Modeling!
\nKaggle Team
", "rawMarkdown": "Hi Kagglers,\n\nWe’re happy to announce the conclusion of the CAFA 6 Protein Function Prediction competition. Thank you everyone for your participation!\n\nThis competition ended with 9,249 registrations and 2,538 participants on 2,172 teams. We had 3,747 submissions from 103 countries. For 631 users (including 31 in the top 100!), this was their first competition. Thank you all for your hard work in this competition and congratulations to our winners and to those who gained a new ranking! \n\nThe top potential winning teams will be contacted via email for the next steps. We look forward to learning more about their winning solutions.\n\nWe've cleaned the leaderboard and disqualified some teams that have violated the rules. If you think you were removed by mistake, or believe you have evidence that suggests another team cheated, please contact [compliance](https://www.kaggle.com/compliance). Please fill in all the fields honestly.\n\nWe highly encourage you to post a solution write-up about your approach and solution in the forums (see [instructions](https://www.kaggle.com/discussions/product-feedback/373153)). You may also refer to [Kaggle Solution Write-Up Documentation](https://www.kaggle.com/solution-write-up-documentation) for guidance. You are also encouraged to [publish your models on Kaggle Models](https://www.kaggle.com/docs/models#publishing-a-model)!\n\nThanks for continuing to make Kaggle a great place to learn, practice, and test data science techniques!\n\nHappy Modeling!\n\nKaggle Team", "votes": null }, { "id": "3514017", "postDate": "08/18/2026 08:48:10", "content": "When will the test set be published, or the competition opened for late submissions?
", "rawMarkdown": "When will the test set be published, or the competition opened for late submissions?", "votes": null }, { "id": "3514507", "postDate": "08/19/2026 15:05:55", "content": "Late submission is not available for this competition. You can find the test set (ground truth set) used in the final evaluation on Kaggle here: https://figshare.com/articles/dataset/CAFA6_Kaggle_Evaluation_Data/32861969
", "rawMarkdown": "Late submission is not available for this competition. You can find the test set (ground truth set) used in the final evaluation on Kaggle here: https://figshare.com/articles/dataset/CAFA6_Kaggle_Evaluation_Data/32861969", "votes": null }, { "id": "3525312", "postDate": "09/17/2026 13:38:51", "content": "Hi! Better late than never :) I finally found a time to clean up my CAFA6 solution and prepare it for LAFA, but I got stuck trying to reproduce the score of my final submission.
\nI downloaded the exact submission.tsv that received 0.42114 on the Kaggle private leaderboard and evaluated it locally using this May 2026 evaluation data and CAFA-evaluator-PK.
\nWith the weighted Fmax scores over the three ontologies within each regime gives:
\nNK = 0.577867\nLK = 0.553833\nPK = 0.228300
\naveraging all 9 ontology × knowledge-regime weighted Fmax scores gives 0.453333
\nI tried both -prop max and -prop fill, and they give exactly the same result…
\nSorry if I missed it and this was already discussed somewhere on the forum. But could you please clarify how to reproduce the private leaderboard score locally, or what differs between the released evaluation setup and the Kaggle private leaderboard evaluation?
\nFor NK/LK I used:
\npython CAFA-evaluator-PK/src/cafaeval/__main__.py \\\n release_kaggle_final_May2026/go-basic.obo \\\n predictions/submission.tsv \\\n release_kaggle_final_May2026/May2026_groundtruth_NK.tsv \\\n -ia release_kaggle_final_May2026/IA.tsv \\\n -no_orphans -norm cafa -prop max \\\n -toi release_kaggle_final_May2026/May2026_groundtruth_terms_of_interest.txt \\\n -max_terms 500 -th_step 0.001\nFor LK I replaced NK file with May2026_groundtruth_LK.tsv. For PK I additionally used:
\n-known release_kaggle_final_May2026/May2026_groundtruth_PK_known.tsv
Hi, all the run commands and arguments are correct, we used -prop fill. For the weighted Fmax scores, here's what I have (locally evaluated) on your submission (50166914.csv) that scored 0.42114 on the leaderboard:
\nFinal average score: 0.42113.
\nIf you used the CAFA-evaluator-PK package, these weighted Fmax scores were taken from column \"f_micro_w\" in \"evaluation_best_f_micro_w.tsv\" result file for each NK/LK/PK setting.
", "rawMarkdown": "Hi, all the run commands and arguments are correct, we used -prop fill. For the weighted Fmax scores, here's what I have (locally evaluated) on your submission (50166914.csv) that scored 0.42114 on the leaderboard: \n- NK: BPO 0.3178, CCO 0.5511, MFO 0.7125. Average NK = 0.5271\n- LK: BPO 0.4285, CCO 0.5634, MFO 0.6131. Average LK = 0.535\n- PK: BPO 0.0935, CCO 0.2657, MFO 0.2446. Average PK = 0.2012\n\nFinal average score: 0.42113.\n\nIf you used the CAFA-evaluator-PK package, these weighted Fmax scores were taken from column \"f_micro_w\" in \"evaluation_best_f_micro_w.tsv\" result file for each NK/LK/PK setting.", "votes": null }, { "id": "3525384", "postDate": "09/17/2026 16:30:16", "content": "Ah, that explains it - I was looking at the wrong output file and using evaluation_best_f_w.tsv. With f_micro_w, everything matches the leaderboard score. Thank you very much, this was extremely helpful!
", "rawMarkdown": "Ah, that explains it - I was looking at the wrong output file and using evaluation_best_f_w.tsv. With f_micro_w, everything matches the leaderboard score. Thank you very much, this was extremely helpful!", "votes": null } ], "comments": [ { "id": 3514017, "author_name": "madshardermller", "author_url": "", "post_date": "08/18/2026 08:48:10", "content": "When will the test set be published, or the competition opened for late submissions?
", "votes": null, "replies": [ { "id": 3514507, "author_name": "ahphan", "author_url": "", "post_date": "08/19/2026 15:05:55", "content": "Late submission is not available for this competition. You can find the test set (ground truth set) used in the final evaluation on Kaggle here: https://figshare.com/articles/dataset/CAFA6_Kaggle_Evaluation_Data/32861969
", "votes": null, "replies": [ { "id": 3525312, "author_name": "antoninadolgorukova", "author_url": "", "post_date": "09/17/2026 13:38:51", "content": "Hi! Better late than never :) I finally found a time to clean up my CAFA6 solution and prepare it for LAFA, but I got stuck trying to reproduce the score of my final submission.
\nI downloaded the exact submission.tsv that received 0.42114 on the Kaggle private leaderboard and evaluated it locally using this May 2026 evaluation data and CAFA-evaluator-PK.
\nWith the weighted Fmax scores over the three ontologies within each regime gives:
\nNK = 0.577867\nLK = 0.553833\nPK = 0.228300
\naveraging all 9 ontology × knowledge-regime weighted Fmax scores gives 0.453333
\nI tried both -prop max and -prop fill, and they give exactly the same result…
\nSorry if I missed it and this was already discussed somewhere on the forum. But could you please clarify how to reproduce the private leaderboard score locally, or what differs between the released evaluation setup and the Kaggle private leaderboard evaluation?
\nFor NK/LK I used:
\npython CAFA-evaluator-PK/src/cafaeval/__main__.py \\\n release_kaggle_final_May2026/go-basic.obo \\\n predictions/submission.tsv \\\n release_kaggle_final_May2026/May2026_groundtruth_NK.tsv \\\n -ia release_kaggle_final_May2026/IA.tsv \\\n -no_orphans -norm cafa -prop max \\\n -toi release_kaggle_final_May2026/May2026_groundtruth_terms_of_interest.txt \\\n -max_terms 500 -th_step 0.001\nFor LK I replaced NK file with May2026_groundtruth_LK.tsv. For PK I additionally used:
\n-known release_kaggle_final_May2026/May2026_groundtruth_PK_known.tsv
Hi, all the run commands and arguments are correct, we used -prop fill. For the weighted Fmax scores, here's what I have (locally evaluated) on your submission (50166914.csv) that scored 0.42114 on the leaderboard:
\nFinal average score: 0.42113.
\nIf you used the CAFA-evaluator-PK package, these weighted Fmax scores were taken from column \"f_micro_w\" in \"evaluation_best_f_micro_w.tsv\" result file for each NK/LK/PK setting.
", "votes": null, "replies": [ { "id": 3525384, "author_name": "antoninadolgorukova", "author_url": "", "post_date": "09/17/2026 16:30:16", "content": "Ah, that explains it - I was looking at the wrong output file and using evaluation_best_f_w.tsv. With f_micro_w, everything matches the leaderboard score. Thank you very much, this was extremely helpful!
", "votes": null, "replies": [] } ] } ] } ] } ] } ], "raw_markdown_by_id": { "3471426": "Hi Kagglers,\n\nWe’re happy to announce the conclusion of the CAFA 6 Protein Function Prediction competition. Thank you everyone for your participation!\n\nThis competition ended with 9,249 registrations and 2,538 participants on 2,172 teams. We had 3,747 submissions from 103 countries. For 631 users (including 31 in the top 100!), this was their first competition. Thank you all for your hard work in this competition and congratulations to our winners and to those who gained a new ranking! \n\nThe top potential winning teams will be contacted via email for the next steps. We look forward to learning more about their winning solutions.\n\nWe've cleaned the leaderboard and disqualified some teams that have violated the rules. If you think you were removed by mistake, or believe you have evidence that suggests another team cheated, please contact [compliance](https://www.kaggle.com/compliance). Please fill in all the fields honestly.\n\nWe highly encourage you to post a solution write-up about your approach and solution in the forums (see [instructions](https://www.kaggle.com/discussions/product-feedback/373153)). You may also refer to [Kaggle Solution Write-Up Documentation](https://www.kaggle.com/solution-write-up-documentation) for guidance. You are also encouraged to [publish your models on Kaggle Models](https://www.kaggle.com/docs/models#publishing-a-model)!\n\nThanks for continuing to make Kaggle a great place to learn, practice, and test data science techniques!\n\nHappy Modeling!\n\nKaggle Team", "3514017": "When will the test set be published, or the competition opened for late submissions?", "3514507": "Late submission is not available for this competition. You can find the test set (ground truth set) used in the final evaluation on Kaggle here: https://figshare.com/articles/dataset/CAFA6_Kaggle_Evaluation_Data/32861969", "3525312": "Hi! Better late than never :) I finally found a time to clean up my CAFA6 solution and prepare it for LAFA, but I got stuck trying to reproduce the score of my final submission.\n\nI downloaded the exact submission.tsv that received 0.42114 on the Kaggle private leaderboard and evaluated it locally using this May 2026 evaluation data and [CAFA-evaluator-PK](https://github.com/claradepaolis/CAFA-evaluator-PK).\n\nWith the weighted Fmax scores over the three ontologies within each regime gives:\n\nNK = 0.577867\nLK = 0.553833\nPK = 0.228300\n\naveraging all 9 ontology × knowledge-regime weighted Fmax scores gives 0.453333[](url)\n\nI tried both -prop max and -prop fill, and they give exactly the same result...\n\nSorry if I missed it and this was already discussed somewhere on the forum. But could you please clarify how to reproduce the private leaderboard score locally, or what differs between the released evaluation setup and the Kaggle private leaderboard evaluation?\n\nFor NK/LK I used:\n```bash\npython CAFA-evaluator-PK/src/cafaeval/__main__.py \\\n release_kaggle_final_May2026/go-basic.obo \\\n predictions/submission.tsv \\\n release_kaggle_final_May2026/May2026_groundtruth_NK.tsv \\\n -ia release_kaggle_final_May2026/IA.tsv \\\n -no_orphans -norm cafa -prop max \\\n -toi release_kaggle_final_May2026/May2026_groundtruth_terms_of_interest.txt \\\n -max_terms 500 -th_step 0.001\n```\n\nFor LK I replaced NK file with May2026_groundtruth_LK.tsv. For PK I additionally used:\n\n`-known release_kaggle_final_May2026/May2026_groundtruth_PK_known.tsv`", "3525331": "Hi, all the run commands and arguments are correct, we used -prop fill. For the weighted Fmax scores, here's what I have (locally evaluated) on your submission (50166914.csv) that scored 0.42114 on the leaderboard: \n- NK: BPO 0.3178, CCO 0.5511, MFO 0.7125. Average NK = 0.5271\n- LK: BPO 0.4285, CCO 0.5634, MFO 0.6131. Average LK = 0.535\n- PK: BPO 0.0935, CCO 0.2657, MFO 0.2446. Average PK = 0.2012\n\nFinal average score: 0.42113.\n\nIf you used the CAFA-evaluator-PK package, these weighted Fmax scores were taken from column \"f_micro_w\" in \"evaluation_best_f_micro_w.tsv\" result file for each NK/LK/PK setting.", "3525384": "Ah, that explains it - I was looking at the wrong output file and using evaluation_best_f_w.tsv. With f_micro_w, everything matches the leaderboard score. Thank you very much, this was extremely helpful!" }, "source": "meta" }