I'm wondering how the train set was selected. In the data description, we can read:
\n\n\nFor the training set, we include all proteins with annotated terms that have been validated by experimental or high-throughput evidence, traceable author statement (evidence code TAS), or inferred by curator (IC). More information about evidence codes can be found here. We use annotations from the UniProtKB release of 2022-11-17.
\n
In my superficial understanding, the curated part of UniProt is Swiss-Prot, so I'd expect the train set to be same as or a subset of Swiss-Prot. But then later on we can read:
\n\n\nMost protein sequences were extracted from the Swiss-Prot database, but a subset of proteins that are not represented in Swiss-Prot were extracted from the TrEMBL database.
\n
Below is the GPT4 overview of both databases:
\nSwiss-Prot and TrEMBL are both components of the UniProt (Universal Protein) knowledgebase, which is a comprehensive resource of protein sequence and annotation data. Here are the differences:\n\nSwiss-Prot: This is the manually annotated and reviewed section of the UniProt database. Each entry in Swiss-Prot has been carefully curated by expert biologists who add a wealth of information to the raw sequence data, such as function, domains, modifications, and disease associations. The high quality of Swiss-Prot data makes it a reliable resource, but its manual curation means that it doesn't include all known proteins and is updated less frequently.\n\nTrEMBL (Translated EMBL): This is the automatically annotated and not reviewed section of the UniProt database. TrEMBL includes protein sequences that are automatically annotated using prediction algorithms and are awaiting manual curation for inclusion into Swiss-Prot. As such, TrEMBL is much larger than Swiss-Prot and is updated more frequently, but its entries may be less reliable due to the lack of manual review.\n\nIn summary, Swiss-Prot is smaller but contains high-quality, manually reviewed data, while TrEMBL is larger and contains automatically annotated data that has not been manually reviewed.\nI'm wondering how the train set was selected. In the data description, we can read:
\n\n\nFor the training set, we include all proteins with annotated terms that have been validated by experimental or high-throughput evidence, traceable author statement (evidence code TAS), or inferred by curator (IC). More information about evidence codes can be found here. We use annotations from the UniProtKB release of 2022-11-17.
\n
In my superficial understanding, the curated part of UniProt is Swiss-Prot, so I'd expect the train set to be same as or a subset of Swiss-Prot. But then later on we can read:
\n\n\nMost protein sequences were extracted from the Swiss-Prot database, but a subset of proteins that are not represented in Swiss-Prot were extracted from the TrEMBL database.
\n
Below is the GPT4 overview of both databases:
\nSwiss-Prot and TrEMBL are both components of the UniProt (Universal Protein) knowledgebase, which is a comprehensive resource of protein sequence and annotation data. Here are the differences:\n\nSwiss-Prot: This is the manually annotated and reviewed section of the UniProt database. Each entry in Swiss-Prot has been carefully curated by expert biologists who add a wealth of information to the raw sequence data, such as function, domains, modifications, and disease associations. The high quality of Swiss-Prot data makes it a reliable resource, but its manual curation means that it doesn't include all known proteins and is updated less frequently.\n\nTrEMBL (Translated EMBL): This is the automatically annotated and not reviewed section of the UniProt database. TrEMBL includes protein sequences that are automatically annotated using prediction algorithms and are awaiting manual curation for inclusion into Swiss-Prot. As such, TrEMBL is much larger than Swiss-Prot and is updated more frequently, but its entries may be less reliable due to the lack of manual review.\n\nIn summary, Swiss-Prot is smaller but contains high-quality, manually reviewed data, while TrEMBL is larger and contains automatically annotated data that has not been manually reviewed.\nI think you might find the answer here: https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/405085#2236274
", "rawMarkdown": "I think you might find the answer here: https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/405085#2236274", "votes": null } ], "comments": [ { "id": 2307043, "author_name": "mtinti", "author_url": "", "post_date": "06/17/2023 19:01:01", "content": "I think you might find the answer here: https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/405085#2236274
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2299868": "I'm wondering how the train set was selected. In the data description, we can read:\n\n>For the training set, we include all proteins with annotated terms that have been validated by experimental or high-throughput evidence, traceable author statement (evidence code TAS), or inferred by curator (IC). More information about evidence codes can be found here. We use annotations from the UniProtKB release of 2022-11-17.\n\nIn my superficial understanding, the curated part of UniProt is Swiss-Prot, so I'd expect the train set to be same as or a subset of Swiss-Prot. But then later on we can read: \n\n>Most protein sequences were extracted from the Swiss-Prot database, but a subset of proteins that are not represented in Swiss-Prot were extracted from the TrEMBL database.\n\nBelow is the GPT4 overview of both databases:\n\n```markdown\nSwiss-Prot and TrEMBL are both components of the UniProt (Universal Protein) knowledgebase, which is a comprehensive resource of protein sequence and annotation data. Here are the differences:\n\nSwiss-Prot: This is the manually annotated and reviewed section of the UniProt database. Each entry in Swiss-Prot has been carefully curated by expert biologists who add a wealth of information to the raw sequence data, such as function, domains, modifications, and disease associations. The high quality of Swiss-Prot data makes it a reliable resource, but its manual curation means that it doesn't include all known proteins and is updated less frequently.\n\nTrEMBL (Translated EMBL): This is the automatically annotated and not reviewed section of the UniProt database. TrEMBL includes protein sequences that are automatically annotated using prediction algorithms and are awaiting manual curation for inclusion into Swiss-Prot. As such, TrEMBL is much larger than Swiss-Prot and is updated more frequently, but its entries may be less reliable due to the lack of manual review.\n\nIn summary, Swiss-Prot is smaller but contains high-quality, manually reviewed data, while TrEMBL is larger and contains automatically annotated data that has not been manually reviewed.\n```", "2307043": "I think you might find the answer here: https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/405085#2236274" }, "source": "meta" }