It looks to me like I find all the go annotations of testsuperset.fasta on uniprot.org. (I checked some random entries). What prevents me of using these to train on more data? Wouldn't that be cheating?
", "messages": [ { "id": "3308641", "postDate": "10/29/2025 20:49:08", "content": "It looks to me like I find all the go annotations of testsuperset.fasta on uniprot.org. (I checked some random entries). What prevents me of using these to train on more data? Wouldn't that be cheating?
", "rawMarkdown": "It looks to me like I find all the go annotations of testsuperset.fasta on uniprot.org. (I checked some random entries). What prevents me of using these to train on more data? Wouldn't that be cheating?", "votes": null }, { "id": "3308642", "postDate": "10/29/2025 20:54:15", "content": "Only new experimentally-validated annotations will be used for final evaluation. New annotations are those that are not in UniProt by the time the competition closes.
", "rawMarkdown": "Only *new* experimentally-validated annotations will be used for final evaluation. New annotations are those that are *not* in UniProt by the time the competition closes.", "votes": null }, { "id": "3308643", "postDate": "10/29/2025 20:59:00", "content": "The test superset includes all proteins (both annotated and un-annotated) in SwissProt that belong to organisms listed in the selected-taxon-list.tsv file. The leaderboard and the final evaluation are carried out in a way that any known/public experimental annotations are excluded, i.e., you will get a score of zero if you only use the existing experimental GO annotations.
The leaderboard ground truth consists of \"hold-out\" experimental annotations that are not in UniProt, and the final evaluation ground truth will consist of novel experimental annotations that accumulate after the submission deadline.
", "rawMarkdown": "The test superset includes all proteins (both annotated and un-annotated) in SwissProt that belong to organisms listed in the `selected-taxon-list.tsv` file. The leaderboard and the final evaluation are carried out in a way that any known/public experimental annotations are excluded, i.e., you will get a score of zero if you only use the existing experimental GO annotations. \n\nThe leaderboard ground truth consists of \"hold-out\" experimental annotations that are not in UniProt, and the final evaluation ground truth will consist of novel experimental annotations that accumulate after the submission deadline.", "votes": null }, { "id": "3311204", "postDate": "11/04/2025 13:29:44", "content": "Are the “new experimentally-validated annotations” those that will later receive one of these evidence codes — EXP, IDA, IPI, IMP, IGI, or IEP — or could other evidence codes also be included in the holdout set?
", "rawMarkdown": "Are the “new experimentally-validated annotations” those that will later receive one of these evidence codes — EXP, IDA, IPI, IMP, IGI, or IEP — or could other evidence codes also be included in the holdout set?", "votes": null }, { "id": "3311230", "postDate": "11/04/2025 14:40:01", "content": "\n\nexperimental or high-throughput evidence, traceable author statement (evidence code TAS), or inferred by curator (IC)
\n
Only these evidence codes (EXP, IDA, IPI, IMP, IGI, IEP, HTP, HDA, HMP, HGI, HEP, IC, TAS) were used in the training data and the leaderboard, and will be used for the final evaluation set. See all GO evidence codes here
", "rawMarkdown": ">experimental or high-throughput evidence, traceable author statement (evidence code TAS), or inferred by curator (IC)\n\nOnly these evidence codes (EXP, IDA, IPI, IMP, IGI, IEP, HTP, HDA, HMP, HGI, HEP, IC, TAS) were used in the training data and the leaderboard, and will be used for the final evaluation set. See all GO evidence codes [here](http://geneontology.org/docs/guide-go-evidence-codes/)", "votes": null }, { "id": "3371226", "postDate": "12/11/2025 11:10:51", "content": "Sorry for being the stupid here, so I am trying to formulate my main question in plain english to anyone who can help..\nhttp://ftp.ebi.ac.uk/pub/databases/GO/goa/UNIPROT/goa_uniprot_all.gaf.gz\nGiven the uniprot GOA, which has the latest release on 10th Dec I think, and if I make external annotations from it, based on train and test set, and use it during hierarchy propagation..
\nIs that ALLOWED or not? according to the rule, as in some place it says we can use external sources, but I count not find restrictions which says \" external sources gone public after date xxxx you not allowed to use\"…
\nthank you for the reply…
", "rawMarkdown": "Sorry for being the stupid here, so I am trying to formulate my main question in plain english to anyone who can help..\nhttp://ftp.ebi.ac.uk/pub/databases/GO/goa/UNIPROT/goa_uniprot_all.gaf.gz\nGiven the uniprot GOA, which has the latest release on 10th Dec I think, and if I make external annotations from it, based on train and test set, and use it during hierarchy propagation..\n\nIs that ALLOWED or not? according to the rule, as in some place it says we can use external sources, but I count not find restrictions which says \" external sources gone public after date xxxx you not allowed to use\"...\n\nthank you for the reply...", "votes": null }, { "id": "3371627", "postDate": "12/11/2025 15:28:47", "content": "You are allowed to use the most recent release of the UniProt GOA. Keep in mind that in the final evaluation, we will exclude all annotations with experimental/curated evidence codes that are made public before submission closes, and only score on new annotations accumulated between submission deadline and final evaluation date.
", "rawMarkdown": "You are allowed to use the most recent release of the UniProt GOA. Keep in mind that in the final evaluation, we will exclude all annotations with experimental/curated evidence codes that are made public before submission closes, and only score on new annotations accumulated between submission deadline and final evaluation date.", "votes": null } ], "comments": [ { "id": 3308642, "author_name": "claradepaolis", "author_url": "", "post_date": "10/29/2025 20:54:15", "content": "Only new experimentally-validated annotations will be used for final evaluation. New annotations are those that are not in UniProt by the time the competition closes.
", "votes": null, "replies": [ { "id": 3311204, "author_name": "xaxipirulazo", "author_url": "", "post_date": "11/04/2025 13:29:44", "content": "Are the “new experimentally-validated annotations” those that will later receive one of these evidence codes — EXP, IDA, IPI, IMP, IGI, or IEP — or could other evidence codes also be included in the holdout set?
", "votes": null, "replies": [ { "id": 3311230, "author_name": "ahphan", "author_url": "", "post_date": "11/04/2025 14:40:01", "content": "\n\nexperimental or high-throughput evidence, traceable author statement (evidence code TAS), or inferred by curator (IC)
\n
Only these evidence codes (EXP, IDA, IPI, IMP, IGI, IEP, HTP, HDA, HMP, HGI, HEP, IC, TAS) were used in the training data and the leaderboard, and will be used for the final evaluation set. See all GO evidence codes here
", "votes": null, "replies": [] } ] } ] }, { "id": 3308643, "author_name": "ahphan", "author_url": "", "post_date": "10/29/2025 20:59:00", "content": "The test superset includes all proteins (both annotated and un-annotated) in SwissProt that belong to organisms listed in the selected-taxon-list.tsv file. The leaderboard and the final evaluation are carried out in a way that any known/public experimental annotations are excluded, i.e., you will get a score of zero if you only use the existing experimental GO annotations.
The leaderboard ground truth consists of \"hold-out\" experimental annotations that are not in UniProt, and the final evaluation ground truth will consist of novel experimental annotations that accumulate after the submission deadline.
", "votes": null, "replies": [] }, { "id": 3371226, "author_name": "viktorbecz", "author_url": "", "post_date": "12/11/2025 11:10:51", "content": "Sorry for being the stupid here, so I am trying to formulate my main question in plain english to anyone who can help..\nhttp://ftp.ebi.ac.uk/pub/databases/GO/goa/UNIPROT/goa_uniprot_all.gaf.gz\nGiven the uniprot GOA, which has the latest release on 10th Dec I think, and if I make external annotations from it, based on train and test set, and use it during hierarchy propagation..
\nIs that ALLOWED or not? according to the rule, as in some place it says we can use external sources, but I count not find restrictions which says \" external sources gone public after date xxxx you not allowed to use\"…
\nthank you for the reply…
", "votes": null, "replies": [ { "id": 3371627, "author_name": "ahphan", "author_url": "", "post_date": "12/11/2025 15:28:47", "content": "You are allowed to use the most recent release of the UniProt GOA. Keep in mind that in the final evaluation, we will exclude all annotations with experimental/curated evidence codes that are made public before submission closes, and only score on new annotations accumulated between submission deadline and final evaluation date.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3308641": "It looks to me like I find all the go annotations of testsuperset.fasta on uniprot.org. (I checked some random entries). What prevents me of using these to train on more data? Wouldn't that be cheating?", "3308642": "Only *new* experimentally-validated annotations will be used for final evaluation. New annotations are those that are *not* in UniProt by the time the competition closes.", "3308643": "The test superset includes all proteins (both annotated and un-annotated) in SwissProt that belong to organisms listed in the `selected-taxon-list.tsv` file. The leaderboard and the final evaluation are carried out in a way that any known/public experimental annotations are excluded, i.e., you will get a score of zero if you only use the existing experimental GO annotations. \n\nThe leaderboard ground truth consists of \"hold-out\" experimental annotations that are not in UniProt, and the final evaluation ground truth will consist of novel experimental annotations that accumulate after the submission deadline.", "3311204": "Are the “new experimentally-validated annotations” those that will later receive one of these evidence codes — EXP, IDA, IPI, IMP, IGI, or IEP — or could other evidence codes also be included in the holdout set?", "3311230": ">experimental or high-throughput evidence, traceable author statement (evidence code TAS), or inferred by curator (IC)\n\nOnly these evidence codes (EXP, IDA, IPI, IMP, IGI, IEP, HTP, HDA, HMP, HGI, HEP, IC, TAS) were used in the training data and the leaderboard, and will be used for the final evaluation set. See all GO evidence codes [here](http://geneontology.org/docs/guide-go-evidence-codes/)", "3371226": "Sorry for being the stupid here, so I am trying to formulate my main question in plain english to anyone who can help..\nhttp://ftp.ebi.ac.uk/pub/databases/GO/goa/UNIPROT/goa_uniprot_all.gaf.gz\nGiven the uniprot GOA, which has the latest release on 10th Dec I think, and if I make external annotations from it, based on train and test set, and use it during hierarchy propagation..\n\nIs that ALLOWED or not? according to the rule, as in some place it says we can use external sources, but I count not find restrictions which says \" external sources gone public after date xxxx you not allowed to use\"...\n\nthank you for the reply...", "3371627": "You are allowed to use the most recent release of the UniProt GOA. Keep in mind that in the final evaluation, we will exclude all annotations with experimental/curated evidence codes that are made public before submission closes, and only score on new annotations accumulated between submission deadline and final evaluation date." }, "source": "meta" }