I have two quick questions regarding the data sets:
\n\nSorry, if I didn't use the correct chemistry terminology.
", "messages": [ { "id": "560692", "postDate": "06/25/2019 17:55:27", "content": "I have two quick questions regarding the data sets:
\n\nSorry, if I didn't use the correct chemistry terminology.
", "rawMarkdown": "I have two quick questions regarding the data sets:\n\n1. train.csv contains the J coupling constants between two atoms. Do the \"j coupling\" edges themselves of a molecule correspond to the chemical bond layout? \n2. structure.csv contains the positions, but no bond information. Do you think that the final test.csv provides us with the entire coupling type information (and if 1. is true, also with the bond information) for each molecule as in train.csv? When I looked at the given test.csv it seemed that way. \n\nSorry, if I didn't use the correct chemistry terminology.", "votes": null }, { "id": "560695", "postDate": "06/25/2019 17:58:56", "content": "My understanding is that a 1JXX row does represent a single bond between two atoms (that's what the \"1\" means), but the same is not true for 2JXX and 3JXX rows. For a 2JXX row, two bonds separate the paired atoms, while there are three bonds between atoms represented in a 3JXX row.
", "rawMarkdown": "My understanding is that a 1JXX row does represent a single bond between two atoms (that's what the \"1\" means), but the same is not true for 2JXX and 3JXX rows. For a 2JXX row, two bonds separate the paired atoms, while there are three bonds between atoms represented in a 3JXX row.", "votes": null }, { "id": "561412", "postDate": "06/26/2019 12:30:46", "content": "Thanks for your reply, so the take away is that we can reconstruct the chemical bonds from the 1JXX couplings. \nNow the question remains, whether we are always given all of the 1JXX couplings for each molecule within in the hidden test set. If that is the case we can reconstruct the bond structure of each molecule.
", "rawMarkdown": "Thanks for your reply, so the take away is that we can reconstruct the chemical bonds from the 1JXX couplings. \nNow the question remains, whether we are always given all of the 1JXX couplings for each molecule within in the hidden test set. If that is the case we can reconstruct the bond structure of each molecule.", "votes": null }, { "id": "561506", "postDate": "06/26/2019 13:53:48", "content": "Hi Max,
\nI don't think it's quite right that we can reconstruct all of the chemical bonds from the 1JXX couplings. There are bonds that don't have a corresponding 1JXX row in the train or test sets. For one thing, the train and test sets only include 1J rows for Hydrogen-Nitrogen (that is, HN) or Hydrogen-Carbon (HC) 1J coupling; but there are bonds between, for example, Oxygen and Hydrogen atoms in some of the molecules. That said, the structure files have information about all atoms in each molecule. You can use this information, along with a tool like OpenBabel, to get at the bond structure of each molecule.
EDIT: Didn't have the data in front of me, couldn't remember if it was 1JHN or 1JHH... Whoops.
", "rawMarkdown": "Hi Max, \nI don't think it's quite right that we can reconstruct *all* of the chemical bonds from the 1JXX couplings. There are bonds that don't have a corresponding 1JXX row in the train or test sets. For one thing, the train and test sets only include 1J rows for Hydrogen-Nitrogen (that is, HN) or Hydrogen-Carbon (HC) 1J coupling; but there are bonds between, for example, Oxygen and Hydrogen atoms in some of the molecules. That said, the structure files have information about all atoms in each molecule. You can use this information, along with a tool like OpenBabel, to get at the bond structure of each molecule.\n\nEDIT: Didn't have the data in front of me, couldn't remember if it was 1JHN or 1JHH... Whoops.", "votes": null }, { "id": "562795", "postDate": "06/27/2019 14:42:31", "content": "J-couplings are only provided for H-* bonds for the public train and test sets, so you cannot reconstruct all the bonds from coupling information only. You would need 1JCC, 1JCN couplings for that as well.\nThere were few kernels that reconstruct bonds using other methods in this thread - https://www.kaggle.com/c/champs-scalar-coupling/discussion/96452
", "rawMarkdown": "J-couplings are only provided for H-* bonds for the public train and test sets, so you cannot reconstruct all the bonds from coupling information only. You would need 1JCC, 1JCN couplings for that as well.\nThere were few kernels that reconstruct bonds using other methods in this thread - https://www.kaggle.com/c/champs-scalar-coupling/discussion/96452", "votes": null } ], "comments": [ { "id": 560695, "author_name": "abendorf", "author_url": "", "post_date": "06/25/2019 17:58:56", "content": "My understanding is that a 1JXX row does represent a single bond between two atoms (that's what the \"1\" means), but the same is not true for 2JXX and 3JXX rows. For a 2JXX row, two bonds separate the paired atoms, while there are three bonds between atoms represented in a 3JXX row.
", "votes": null, "replies": [] }, { "id": 561412, "author_name": "le1nux", "author_url": "", "post_date": "06/26/2019 12:30:46", "content": "Thanks for your reply, so the take away is that we can reconstruct the chemical bonds from the 1JXX couplings. \nNow the question remains, whether we are always given all of the 1JXX couplings for each molecule within in the hidden test set. If that is the case we can reconstruct the bond structure of each molecule.
", "votes": null, "replies": [ { "id": 561506, "author_name": "abendorf", "author_url": "", "post_date": "06/26/2019 13:53:48", "content": "Hi Max,
\nI don't think it's quite right that we can reconstruct all of the chemical bonds from the 1JXX couplings. There are bonds that don't have a corresponding 1JXX row in the train or test sets. For one thing, the train and test sets only include 1J rows for Hydrogen-Nitrogen (that is, HN) or Hydrogen-Carbon (HC) 1J coupling; but there are bonds between, for example, Oxygen and Hydrogen atoms in some of the molecules. That said, the structure files have information about all atoms in each molecule. You can use this information, along with a tool like OpenBabel, to get at the bond structure of each molecule.
EDIT: Didn't have the data in front of me, couldn't remember if it was 1JHN or 1JHH... Whoops.
", "votes": null, "replies": [] }, { "id": 562795, "author_name": "vzaguskin", "author_url": "", "post_date": "06/27/2019 14:42:31", "content": "J-couplings are only provided for H-* bonds for the public train and test sets, so you cannot reconstruct all the bonds from coupling information only. You would need 1JCC, 1JCN couplings for that as well.\nThere were few kernels that reconstruct bonds using other methods in this thread - https://www.kaggle.com/c/champs-scalar-coupling/discussion/96452
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "560692": "I have two quick questions regarding the data sets:\n\n1. train.csv contains the J coupling constants between two atoms. Do the \"j coupling\" edges themselves of a molecule correspond to the chemical bond layout? \n2. structure.csv contains the positions, but no bond information. Do you think that the final test.csv provides us with the entire coupling type information (and if 1. is true, also with the bond information) for each molecule as in train.csv? When I looked at the given test.csv it seemed that way. \n\nSorry, if I didn't use the correct chemistry terminology.", "560695": "My understanding is that a 1JXX row does represent a single bond between two atoms (that's what the \"1\" means), but the same is not true for 2JXX and 3JXX rows. For a 2JXX row, two bonds separate the paired atoms, while there are three bonds between atoms represented in a 3JXX row.", "561412": "Thanks for your reply, so the take away is that we can reconstruct the chemical bonds from the 1JXX couplings. \nNow the question remains, whether we are always given all of the 1JXX couplings for each molecule within in the hidden test set. If that is the case we can reconstruct the bond structure of each molecule.", "561506": "Hi Max, \nI don't think it's quite right that we can reconstruct *all* of the chemical bonds from the 1JXX couplings. There are bonds that don't have a corresponding 1JXX row in the train or test sets. For one thing, the train and test sets only include 1J rows for Hydrogen-Nitrogen (that is, HN) or Hydrogen-Carbon (HC) 1J coupling; but there are bonds between, for example, Oxygen and Hydrogen atoms in some of the molecules. That said, the structure files have information about all atoms in each molecule. You can use this information, along with a tool like OpenBabel, to get at the bond structure of each molecule.\n\nEDIT: Didn't have the data in front of me, couldn't remember if it was 1JHN or 1JHH... Whoops.", "562795": "J-couplings are only provided for H-* bonds for the public train and test sets, so you cannot reconstruct all the bonds from coupling information only. You would need 1JCC, 1JCN couplings for that as well.\nThere were few kernels that reconstruct bonds using other methods in this thread - https://www.kaggle.com/c/champs-scalar-coupling/discussion/96452" }, "source": "meta" }