Welcome to the BMS Molecular Translation Competition, we are excited to pose this challenge to the machine learning community!
\nApplying machine learning approaches to scientific challenges has tremendous potential. For chemical research, a recent review article of optical chemical structure recognition tools highlights the need for advancing the technology. We're looking forward to what the community can offer!
\nGood Luck!
", "messages": [ { "id": 1224592, "postDate": "2021-03-02T22:03:05.677Z", "content": "Welcome to the BMS Molecular Translation Competition, we are excited to pose this challenge to the machine learning community!
\nApplying machine learning approaches to scientific challenges has tremendous potential. For chemical research, a recent review article of optical chemical structure recognition tools highlights the need for advancing the technology. We're looking forward to what the community can offer!
\nGood Luck!
", "rawMarkdown": "Welcome to the BMS Molecular Translation Competition, we are excited to pose this challenge to the machine learning community! \n\nApplying machine learning approaches to scientific challenges has tremendous potential. For chemical research, a recent [review article](https://doi.org/10.1186/s13321-020-00465-0) of optical chemical structure recognition tools highlights the need for advancing the technology. We're looking forward to what the community can offer!\n\nGood Luck!", "votes": 60 }, { "id": 1224685, "postDate": "2021-03-03T01:25:01.270Z", "content": "@jakealbrecht1337 Thank you for organizing this.
\nI'm currently working on a notebook that would preprocess the data and save it as a dataset, similar to what I did here. I'm planning to publish it as a public dataset, but I will link it to this competition and put a disclaimer about accepting the rules of the competition in the description.
\nPlease let me know if you would prefer not to see this and I will remove the dataset.
", "rawMarkdown": "@jakealbrecht1337 Thank you for organizing this. \n\nI'm currently working on a notebook that would preprocess the data and save it as a dataset, similar to what I did [here](https://www.kaggle.com/xhlulu/vinbigdata-process-and-resize-to-image). I'm planning to publish it as a public dataset, but I will link it to this competition and put a disclaimer about accepting the rules of the competition in the description.\n\nPlease let me know if you would prefer not to see this and I will remove the dataset.", "votes": 3, "replies": [ { "id": 1225043, "postDate": "2021-03-03T09:07:30.577Z", "content": "I'm interested in hearing which image pre-processing techniques you intend to use.
", "rawMarkdown": "I'm interested in hearing which image pre-processing techniques you intend to use." }, { "id": 1225890, "postDate": "2021-03-04T03:29:16.970Z", "content": "I'm planning to mostly resize, convert to jpg and add padding or stretching so they have a consistent size.
", "rawMarkdown": "I'm planning to mostly resize, convert to jpg and add padding or stretching so they have a consistent size." } ] }, { "id": 1233852, "postDate": "2021-03-10T18:13:25.893Z", "content": "What is missing in this competition is a clear explanation (or links to that) of how to go from the molecular graph to the InChI. We need to know how the atoms are ordered/numbered in order to come up with a unique InChI.
\nThe atoms ordering logic is not described in the given wikipedia page.
@jakealbrecht1337 I have 2 quick questions. After the competition, is the organizer going to unblind all the 1.6M labels in the test set? Second, is there any plan to wrap up the knowledge we learn from this competition and write a paper about it? Thanks!
", "rawMarkdown": "@jakealbrecht1337 I have 2 quick questions. After the competition, is the organizer going to unblind all the 1.6M labels in the test set? Second, is there any plan to wrap up the knowledge we learn from this competition and write a paper about it? Thanks!", "votes": 1, "replies": [ { "id": 1293247, "postDate": "2021-05-04T17:29:03.350Z", "content": "I think the idea of writing a paper on the result is excellent, that is something we will definitely explore after things wrap up! Keeping with Kaggle convention, we are leaning toward keeping the test labels blinded to give teams a chance to test winning solutions without the temptation to look up the answers.
", "rawMarkdown": "I think the idea of writing a paper on the result is excellent, that is something we will definitely explore after things wrap up! Keeping with Kaggle convention, we are leaning toward keeping the test labels blinded to give teams a chance to test winning solutions without the temptation to look up the answers.", "votes": 4 }, { "id": 1293780, "postDate": "2021-05-05T06:54:28.583Z", "content": "Writing a paper is very interesting. I've also worked on papers with organizers at previous chemistry-related competitions I got good results.
\nI hope I can work writing a paper this time like before.
", "rawMarkdown": "Writing a paper is very interesting. I've also worked on papers with organizers at previous chemistry-related competitions I got good results.\n\nI hope I can work writing a paper this time like before.", "votes": 1 } ] }, { "id": 1225001, "postDate": "2021-03-03T08:24:32.243Z", "content": "Fantastic challenge, thank you so much for putting this on!
\nOne thing I wanted to clarify: is the private test set included in the test folder we are given and just not scored on the LB until the end? I don't see how else it would get scored
", "rawMarkdown": "Fantastic challenge, thank you so much for putting this on!\n\nOne thing I wanted to clarify: is the private test set included in the test folder we are given and just not scored on the LB until the end? I don't see how else it would get scored", "votes": 1, "replies": [ { "id": 1228791, "postDate": "2021-03-06T19:07:45.280Z", "content": "tagging @jakealbrecht1337 @addisonhoward @inversion for visibility, thank you in advance
", "rawMarkdown": "tagging @jakealbrecht1337 @addisonhoward @inversion for visibility, thank you in advance" }, { "id": 1230071, "postDate": "2021-03-07T18:49:56.757Z", "content": "Correct. You make both Public and Private predictions in your submissions. You see you Public scores upon submission, but the Private score is only revealed at the end of the competition.
", "rawMarkdown": "Correct. You make both Public and Private predictions in your submissions. You see you Public scores upon submission, but the Private score is only revealed at the end of the competition.", "votes": 2 }, { "id": 1230075, "postDate": "2021-03-07T18:52:04.767Z", "content": "Thank you for the clarification, I understand it now @inversion
", "rawMarkdown": "Thank you for the clarification, I understand it now @inversion " } ] }, { "id": 1224651, "postDate": "2021-03-03T00:06:57.887Z", "content": "Wow! This looks very challenging. Do you have any recommended papers we can see previous approaches?
", "rawMarkdown": "Wow! This looks very challenging. Do you have any recommended papers we can see previous approaches?", "votes": 1, "replies": [ { "id": 1316081, "postDate": "2021-05-20T09:37:56.777Z", "content": "Here are two articles that should help:
\nhttps://jcheminf.biomedcentral.com/articles/10.1186/s13321-015-0068-4
\n
\nhttps://jcheminf.biomedcentral.com/articles/10.1186/s13321-020-00465-0
@jakealbrecht1337 Hi there, I believe some of the molecules are mislabelled as searching for the inchi key on chemspider returns a null result, an example of this is sample 3 for the train set, I imagine it is a low percentage of the train set that is as such
", "rawMarkdown": "@jakealbrecht1337 Hi there, I believe some of the molecules are mislabelled as searching for the inchi key on chemspider returns a null result, an example of this is sample 3 for the train set, I imagine it is a low percentage of the train set that is as such ", "votes": 2 }, { "id": 1224747, "postDate": "2021-03-03T03:07:47.427Z", "content": "I'm curious if @inversion will be participating :D He has a strong chance of winning 😁
", "rawMarkdown": "I'm curious if @inversion will be participating :D He has a strong chance of winning 😁", "votes": 2, "replies": [ { "id": 1224808, "postDate": "2021-03-03T04:48:22.357Z", "content": "@inversion prepared the competition on the Kaggle side! He's the reason we were able to launch this!
", "rawMarkdown": "@inversion prepared the competition on the Kaggle side! He's the reason we were able to launch this!", "votes": 6 } ] }, { "id": 1224832, "postDate": "2021-03-03T05:16:09.917Z", "content": "how to solve this problem
", "rawMarkdown": "how to solve this problem", "votes": -2 }, { "id": 1340436, "postDate": "2021-06-07T23:38:44.680Z", "content": "So far, we have not received any emails about how to get the first prize.Therefore, we have to seek answers here.If you can respond to this question, we would be very grateful.
", "rawMarkdown": "So far, we have not received any emails about how to get the first prize.Therefore, we have to seek answers here.If you can respond to this question, we would be very grateful.", "replies": [ { "id": 1340546, "postDate": "2021-06-08T04:42:22.077Z", "content": "Hi there,
\nWe are continuing our investigation of the top 3 winners and will reach out when those have been finalized. Thanks for your patience as we continue to finalize the leaderboard. You will receive an email from me shortly :)
", "rawMarkdown": "Hi there,\n\nWe are continuing our investigation of the top 3 winners and will reach out when those have been finalized. Thanks for your patience as we continue to finalize the leaderboard. You will receive an email from me shortly :)", "votes": 1 }, { "id": 1342417, "postDate": "2021-06-09T12:33:37.387Z", "content": "@zhongfeisheng6666 Congrats! Everyone is waiting your solution too:)
", "rawMarkdown": "@zhongfeisheng6666 Congrats! Everyone is waiting your solution too:)", "votes": 1 } ] }, { "id": 1335065, "postDate": "2021-06-04T02:01:04.230Z", "content": "Sorry to bother you . Our team submitted 1.34 to be 34, and then suddenly we can’t see our ranking，Can someone help us solve this problem？
", "rawMarkdown": "Sorry to bother you . Our team submitted 1.34 to be 34, and then suddenly we can’t see our ranking，Can someone help us solve this problem？" }, { "id": 1234766, "postDate": "2021-03-11T15:07:23.580Z", "content": "Hi everybody! I'm excited to be part of this competition and to have the opportunity to play around with this interesting dataset! 😃
", "rawMarkdown": "Hi everybody! I'm excited to be part of this competition and to have the opportunity to play around with this interesting dataset! 😃" }, { "id": 1228522, "postDate": "2021-03-06T14:01:37.513Z", "content": "How can the 2D projection restore 3D object? Of course, you can draw a scetch from InChI formula, but the reverse task is not fully feasible. You can restore the first part of the formula, the composition of the molecule. Second part, connections depend on how to score graph nodes. Which node of the graph to count first?
", "rawMarkdown": "How can the 2D projection restore 3D object? Of course, you can draw a scetch from InChI formula, but the reverse task is not fully feasible. You can restore the first part of the formula, the composition of the molecule. Second part, connections depend on how to score graph nodes. Which node of the graph to count first?", "replies": [ { "id": 1228790, "postDate": "2021-03-06T19:06:45.933Z", "content": "I believe it is possible to restore the exact InChI from a 2D projection. \"Which node of the graph to count first\" is pretty complicated though. The numbering should be based on IUPAC rules. So first you have to find the longest carbon chain, then find the connecting moieties, and so on. Therefor, adapting existing code for this task will probably be the best route. RDKit has MolToInchi so if you're able to load/build your molecule in RDKit, you can just get the InChI using that.
Sound like an amazing challenge! Thanks for organizing it
", "rawMarkdown": "Sound like an amazing challenge! Thanks for organizing it" }, { "id": 1225860, "postDate": "2021-03-04T02:34:10.470Z", "content": "Are there any good guides on how to write and understand the InChI notation?
", "rawMarkdown": "Are there any good guides on how to write and understand the InChI notation?", "replies": [ { "id": 1225864, "postDate": "2021-03-04T02:46:00.930Z", "content": "For now, the best resource I have found is this.
\nThe FAQ has some interesting information. There are also some download links for the official software.
There's been a couple posts about it already:
\n\nFrom my understanding, the goal of InChI is to accurately describe a unique molecule, making it less intuitive to understand than SMILES.
", "rawMarkdown": "There's been a couple posts about it already:\n- [[Paper]: About InChI chemical structure](https://www.kaggle.com/c/bms-molecular-translation/discussion/223457)\n- [Why InChl?](https://www.kaggle.com/c/bms-molecular-translation/discussion/223213)\n\nFrom my understanding, the goal of InChI is to accurately describe a unique molecule, making it less intuitive to understand than SMILES.", "votes": 3 }, { "id": 1236321, "postDate": "2021-03-13T02:24:19.950Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 1227773, "postDate": "2021-03-05T20:58:06.807Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2010883, "postDate": "2022-10-31T08:02:46.517Z", "content": "Thank for your sharing
", "rawMarkdown": "Thank for your sharing" }, { "id": 1228901, "postDate": "2021-03-06T22:02:20.467Z", "content": "Thank you for this competition!
", "rawMarkdown": "Thank you for this competition!" } ], "comments": [ { "id": 1224685, "author_name": "xhlulu", "author_url": "", "post_date": "2021-03-03T01:25:01.270000", "content": "@jakealbrecht1337 Thank you for organizing this.
\nI'm currently working on a notebook that would preprocess the data and save it as a dataset, similar to what I did here. I'm planning to publish it as a public dataset, but I will link it to this competition and put a disclaimer about accepting the rules of the competition in the description.
\nPlease let me know if you would prefer not to see this and I will remove the dataset.
", "votes": 3, "replies": [ { "id": 1225043, "author_name": "Michael99999", "author_url": "", "post_date": "2021-03-03T09:07:30.577000", "content": "I'm interested in hearing which image pre-processing techniques you intend to use.
", "votes": 0, "replies": [] }, { "id": 1225890, "author_name": "xhlulu", "author_url": "", "post_date": "2021-03-04T03:29:16.970000", "content": "I'm planning to mostly resize, convert to jpg and add padding or stretching so they have a consistent size.
", "votes": 0, "replies": [] } ] }, { "id": 1233852, "author_name": "ISMAX", "author_url": "", "post_date": "2021-03-10T18:13:25.893000", "content": "What is missing in this competition is a clear explanation (or links to that) of how to go from the molecular graph to the InChI. We need to know how the atoms are ordered/numbered in order to come up with a unique InChI.
\nThe atoms ordering logic is not described in the given wikipedia page.
@jakealbrecht1337 I have 2 quick questions. After the competition, is the organizer going to unblind all the 1.6M labels in the test set? Second, is there any plan to wrap up the knowledge we learn from this competition and write a paper about it? Thanks!
", "votes": 1, "replies": [ { "id": 1293247, "author_name": "Jacob Albrecht", "author_url": "", "post_date": "2021-05-04T17:29:03.350000", "content": "I think the idea of writing a paper on the result is excellent, that is something we will definitely explore after things wrap up! Keeping with Kaggle convention, we are leaning toward keeping the test labels blinded to give teams a chance to test winning solutions without the temptation to look up the answers.
", "votes": 4, "replies": [] }, { "id": 1293780, "author_name": "Youhan Lee", "author_url": "", "post_date": "2021-05-05T06:54:28.583000", "content": "Writing a paper is very interesting. I've also worked on papers with organizers at previous chemistry-related competitions I got good results.
\nI hope I can work writing a paper this time like before.
", "votes": 1, "replies": [] } ] }, { "id": 1225001, "author_name": "Matt", "author_url": "", "post_date": "2021-03-03T08:24:32.243000", "content": "Fantastic challenge, thank you so much for putting this on!
\nOne thing I wanted to clarify: is the private test set included in the test folder we are given and just not scored on the LB until the end? I don't see how else it would get scored
", "votes": 1, "replies": [ { "id": 1228791, "author_name": "Matt", "author_url": "", "post_date": "2021-03-06T19:07:45.280000", "content": "tagging @jakealbrecht1337 @addisonhoward @inversion for visibility, thank you in advance
", "votes": 0, "replies": [] }, { "id": 1230071, "author_name": "inversion", "author_url": "", "post_date": "2021-03-07T18:49:56.757000", "content": "Correct. You make both Public and Private predictions in your submissions. You see you Public scores upon submission, but the Private score is only revealed at the end of the competition.
", "votes": 2, "replies": [] }, { "id": 1230075, "author_name": "Matt", "author_url": "", "post_date": "2021-03-07T18:52:04.767000", "content": "Thank you for the clarification, I understand it now @inversion
", "votes": 0, "replies": [] } ] }, { "id": 1224651, "author_name": "CoreyJamesLevinson", "author_url": "", "post_date": "2021-03-03T00:06:57.887000", "content": "Wow! This looks very challenging. Do you have any recommended papers we can see previous approaches?
", "votes": 1, "replies": [ { "id": 1316081, "author_name": "Koubouratou IDJATON", "author_url": "", "post_date": "2021-05-20T09:37:56.777000", "content": "Here are two articles that should help:
\nhttps://jcheminf.biomedcentral.com/articles/10.1186/s13321-015-0068-4
\n
\nhttps://jcheminf.biomedcentral.com/articles/10.1186/s13321-020-00465-0
@jakealbrecht1337 Hi there, I believe some of the molecules are mislabelled as searching for the inchi key on chemspider returns a null result, an example of this is sample 3 for the train set, I imagine it is a low percentage of the train set that is as such
", "votes": 2, "replies": [] }, { "id": 1224747, "author_name": "Sanyam Bhutani", "author_url": "", "post_date": "2021-03-03T03:07:47.427000", "content": "I'm curious if @inversion will be participating :D He has a strong chance of winning 😁
", "votes": 2, "replies": [ { "id": 1224808, "author_name": "Addison Howard", "author_url": "", "post_date": "2021-03-03T04:48:22.357000", "content": "@inversion prepared the competition on the Kaggle side! He's the reason we were able to launch this!
", "votes": 6, "replies": [] } ] }, { "id": 1224832, "author_name": "Abdul Waheed Official", "author_url": "", "post_date": "2021-03-03T05:16:09.917000", "content": "how to solve this problem
", "votes": -2, "replies": [] }, { "id": 1340436, "author_name": "peter", "author_url": "", "post_date": "2021-06-07T23:38:44.680000", "content": "So far, we have not received any emails about how to get the first prize.Therefore, we have to seek answers here.If you can respond to this question, we would be very grateful.
", "votes": 0, "replies": [ { "id": 1340546, "author_name": "Addison Howard", "author_url": "", "post_date": "2021-06-08T04:42:22.077000", "content": "Hi there,
\nWe are continuing our investigation of the top 3 winners and will reach out when those have been finalized. Thanks for your patience as we continue to finalize the leaderboard. You will receive an email from me shortly :)
", "votes": 1, "replies": [] }, { "id": 1342417, "author_name": "Camaro", "author_url": "", "post_date": "2021-06-09T12:33:37.387000", "content": "@zhongfeisheng6666 Congrats! Everyone is waiting your solution too:)
", "votes": 1, "replies": [] } ] }, { "id": 1335065, "author_name": "BarryAllen01", "author_url": "", "post_date": "2021-06-04T02:01:04.230000", "content": "Sorry to bother you . Our team submitted 1.34 to be 34, and then suddenly we can’t see our ranking，Can someone help us solve this problem？
", "votes": 0, "replies": [] }, { "id": 1234766, "author_name": "Anjalytics", "author_url": "", "post_date": "2021-03-11T15:07:23.580000", "content": "Hi everybody! I'm excited to be part of this competition and to have the opportunity to play around with this interesting dataset! 😃
", "votes": 0, "replies": [] }, { "id": 1228522, "author_name": "Nail Fayzullin", "author_url": "", "post_date": "2021-03-06T14:01:37.513000", "content": "How can the 2D projection restore 3D object? Of course, you can draw a scetch from InChI formula, but the reverse task is not fully feasible. You can restore the first part of the formula, the composition of the molecule. Second part, connections depend on how to score graph nodes. Which node of the graph to count first?
", "votes": 0, "replies": [ { "id": 1228790, "author_name": "Matt", "author_url": "", "post_date": "2021-03-06T19:06:45.933000", "content": "I believe it is possible to restore the exact InChI from a 2D projection. \"Which node of the graph to count first\" is pretty complicated though. The numbering should be based on IUPAC rules. So first you have to find the longest carbon chain, then find the connecting moieties, and so on. Therefor, adapting existing code for this task will probably be the best route. RDKit has MolToInchi so if you're able to load/build your molecule in RDKit, you can just get the InChI using that.
Sound like an amazing challenge! Thanks for organizing it
", "votes": 0, "replies": [] }, { "id": 1225860, "author_name": "Bruno", "author_url": "", "post_date": "2021-03-04T02:34:10.470000", "content": "Are there any good guides on how to write and understand the InChI notation?
", "votes": 0, "replies": [ { "id": 1225864, "author_name": "Bruno", "author_url": "", "post_date": "2021-03-04T02:46:00.930000", "content": "For now, the best resource I have found is this.
\nThe FAQ has some interesting information. There are also some download links for the official software.
There's been a couple posts about it already:
\n\nFrom my understanding, the goal of InChI is to accurately describe a unique molecule, making it less intuitive to understand than SMILES.
", "votes": 3, "replies": [] }, { "id": 1236321, "author_name": "", "author_url": "", "post_date": "2021-03-13T02:24:19.950000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1227773, "author_name": "", "author_url": "", "post_date": "2021-03-05T20:58:06.807000", "content": "", "votes": 0, "replies": [] }, { "id": 2010883, "author_name": "Isabella", "author_url": "", "post_date": "2022-10-31T08:02:46.517000", "content": "Thank for your sharing
", "votes": 0, "replies": [] }, { "id": 1228901, "author_name": "Marcos", "author_url": "", "post_date": "2021-03-06T22:02:20.467000", "content": "Thank you for this competition!
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1224592": "Welcome to the BMS Molecular Translation Competition, we are excited to pose this challenge to the machine learning community! \n\nApplying machine learning approaches to scientific challenges has tremendous potential. For chemical research, a recent [review article](https://doi.org/10.1186/s13321-020-00465-0) of optical chemical structure recognition tools highlights the need for advancing the technology. We're looking forward to what the community can offer!\n\nGood Luck!", "1224685": "@jakealbrecht1337 Thank you for organizing this. \n\nI'm currently working on a notebook that would preprocess the data and save it as a dataset, similar to what I did [here](https://www.kaggle.com/xhlulu/vinbigdata-process-and-resize-to-image). I'm planning to publish it as a public dataset, but I will link it to this competition and put a disclaimer about accepting the rules of the competition in the description.\n\nPlease let me know if you would prefer not to see this and I will remove the dataset.", "1233852": "What is missing in this competition is a clear explanation (or links to that) of how to go from the molecular graph to the InChI. We need to know how the atoms are ordered/numbered in order to come up with a unique InChI.\nThe atoms ordering logic is not described in the given wikipedia page.", "1285389": "@jakealbrecht1337 I have 2 quick questions. After the competition, is the organizer going to unblind all the 1.6M labels in the test set? Second, is there any plan to wrap up the knowledge we learn from this competition and write a paper about it? Thanks!", "1225001": "Fantastic challenge, thank you so much for putting this on!\n\nOne thing I wanted to clarify: is the private test set included in the test folder we are given and just not scored on the LB until the end? I don't see how else it would get scored", "1224651": "Wow! This looks very challenging. Do you have any recommended papers we can see previous approaches?", "1233463": "@jakealbrecht1337 Hi there, I believe some of the molecules are mislabelled as searching for the inchi key on chemspider returns a null result, an example of this is sample 3 for the train set, I imagine it is a low percentage of the train set that is as such ", "1224747": "I'm curious if @inversion will be participating :D He has a strong chance of winning 😁", "1224832": "how to solve this problem", "1340436": "So far, we have not received any emails about how to get the first prize.Therefore, we have to seek answers here.If you can respond to this question, we would be very grateful.", "1335065": "Sorry to bother you . Our team submitted 1.34 to be 34, and then suddenly we can’t see our ranking，Can someone help us solve this problem？", "1234766": "Hi everybody! I'm excited to be part of this competition and to have the opportunity to play around with this interesting dataset! 😃", "1228522": "How can the 2D projection restore 3D object? Of course, you can draw a scetch from InChI formula, but the reverse task is not fully feasible. You can restore the first part of the formula, the composition of the molecule. Second part, connections depend on how to score graph nodes. Which node of the graph to count first?", "1226801": "Sound like an amazing challenge! Thanks for organizing it", "1225860": "Are there any good guides on how to write and understand the InChI notation?", "1227773": "", "2010883": "Thank for your sharing", "1228901": "Thank you for this competition!" } }