As I understand the green images correspond to the imaging response of a particular protein(s). Judging from the training set, it does not appear to be always the same protein or imaging method. Why is this information not part of the input data? Unless there is something I don't understand, it should have an enormous predictive power and it should not be hard to keep in the record.
", "messages": [ { "id": 1197051, "postDate": "2021-02-11T22:05:18.847Z", "content": "As I understand the green images correspond to the imaging response of a particular protein(s). Judging from the training set, it does not appear to be always the same protein or imaging method. Why is this information not part of the input data? Unless there is something I don't understand, it should have an enormous predictive power and it should not be hard to keep in the record.
", "rawMarkdown": "As I understand the green images correspond to the imaging response of a particular protein(s). Judging from the training set, it does not appear to be always the same protein or imaging method. Why is this information not part of the input data? Unless there is something I don't understand, it should have an enormous predictive power and it should not be hard to keep in the record.", "votes": 3 }, { "id": 1198370, "postDate": "2021-02-13T01:13:56.530Z", "content": "https://www.proteinatlas.org/humanproteome/cell
\nImaging method is consistent. Protein is not consistent.
\nRead more about how its done - what things mean, etc. The above link is a good starting point.
", "rawMarkdown": "[https://www.proteinatlas.org/humanproteome/cell](https://www.proteinatlas.org/humanproteome/cell)\n\nImaging method is consistent. Protein is not consistent.\n\nRead more about how its done - what things mean, etc. The above link is a good starting point. \n \n\n", "replies": [ { "id": 1205667, "postDate": "2021-02-16T21:39:13.920Z", "content": "That's what I suspected, different protein. My point is that I don't understand why that information is not included as an input. In my understanding, it has the potential of improving the prediction significantly and I would be very surprised if those images are recorded without keeping what kind of protein was being imaged.
", "rawMarkdown": "That's what I suspected, different protein. My point is that I don't understand why that information is not included as an input. In my understanding, it has the potential of improving the prediction significantly and I would be very surprised if those images are recorded without keeping what kind of protein was being imaged." }, { "id": 1205805, "postDate": "2021-02-17T02:33:36.910Z", "content": "My basic assumption is that tomorrow someone will generate a new image using a new protein and would love to have single cells in the image identified. The host for this competition has determined what input and what output they want. IMO this particular competition is complex enough already having to segment images and train/predict on 4 channels - adding tabular input to the model would likely be a bridge too far.
\nIf you participate in bunches of these competitions over several years as I have you will realize that there is ALWAYS additional information that could be supplied that will help the model. So I have grown to accept the task of solving the problem given, rather than inventing a problem I want to solve.
\nWent to High School and College in the Denver area - already have a burial plot in Fort Logan National Cemetery - so always interested in potential teaming up with someone from home. Your questions suggests your likely a liberal Democrat but regardless hit me up with a kaggle email if your interested in teaming up for this competition.
", "rawMarkdown": "My basic assumption is that tomorrow someone will generate a new image using a new protein and would love to have single cells in the image identified. The host for this competition has determined what input and what output they want. IMO this particular competition is complex enough already having to segment images and train/predict on 4 channels - adding tabular input to the model would likely be a bridge too far.\n\nIf you participate in bunches of these competitions over several years as I have you will realize that there is ALWAYS additional information that could be supplied that will help the model. So I have grown to accept the task of solving the problem given, rather than inventing a problem I want to solve.\n\nWent to High School and College in the Denver area - already have a burial plot in Fort Logan National Cemetery - so always interested in potential teaming up with someone from home. Your questions suggests your likely a liberal Democrat but regardless hit me up with a kaggle email if your interested in teaming up for this competition." }, { "id": 1210586, "postDate": "2021-02-19T14:35:24.583Z", "content": "@fcueto It's important to keep in mind that we do this as part of discovering and understanding unknown proteins in laboratory research. While the protein information could potentially be used as input to better performance in this challenge, it would not give us better performance in the instances where we are hoping to be able to use the models from this challenge as the information may not always be available.
", "rawMarkdown": "@fcueto It's important to keep in mind that we do this as part of discovering and understanding unknown proteins in laboratory research. While the protein information could potentially be used as input to better performance in this challenge, it would not give us better performance in the instances where we are hoping to be able to use the models from this challenge as the information may not always be available." } ] } ], "comments": [ { "id": 1198370, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2021-02-13T01:13:56.530000", "content": "https://www.proteinatlas.org/humanproteome/cell
\nImaging method is consistent. Protein is not consistent.
\nRead more about how its done - what things mean, etc. The above link is a good starting point.
", "votes": 0, "replies": [ { "id": 1205667, "author_name": "Malkovrulo", "author_url": "", "post_date": "2021-02-16T21:39:13.920000", "content": "That's what I suspected, different protein. My point is that I don't understand why that information is not included as an input. In my understanding, it has the potential of improving the prediction significantly and I would be very surprised if those images are recorded without keeping what kind of protein was being imaged.
", "votes": 0, "replies": [] }, { "id": 1205805, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2021-02-17T02:33:36.910000", "content": "My basic assumption is that tomorrow someone will generate a new image using a new protein and would love to have single cells in the image identified. The host for this competition has determined what input and what output they want. IMO this particular competition is complex enough already having to segment images and train/predict on 4 channels - adding tabular input to the model would likely be a bridge too far.
\nIf you participate in bunches of these competitions over several years as I have you will realize that there is ALWAYS additional information that could be supplied that will help the model. So I have grown to accept the task of solving the problem given, rather than inventing a problem I want to solve.
\nWent to High School and College in the Denver area - already have a burial plot in Fort Logan National Cemetery - so always interested in potential teaming up with someone from home. Your questions suggests your likely a liberal Democrat but regardless hit me up with a kaggle email if your interested in teaming up for this competition.
", "votes": 0, "replies": [] }, { "id": 1210586, "author_name": "Casper Winsnes", "author_url": "", "post_date": "2021-02-19T14:35:24.583000", "content": "@fcueto It's important to keep in mind that we do this as part of discovering and understanding unknown proteins in laboratory research. While the protein information could potentially be used as input to better performance in this challenge, it would not give us better performance in the instances where we are hoping to be able to use the models from this challenge as the information may not always be available.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1197051": "As I understand the green images correspond to the imaging response of a particular protein(s). Judging from the training set, it does not appear to be always the same protein or imaging method. Why is this information not part of the input data? Unless there is something I don't understand, it should have an enormous predictive power and it should not be hard to keep in the record.", "1198370": "[https://www.proteinatlas.org/humanproteome/cell](https://www.proteinatlas.org/humanproteome/cell)\n\nImaging method is consistent. Protein is not consistent.\n\nRead more about how its done - what things mean, etc. The above link is a good starting point. \n \n\n" } }