Hello,
\n\nI have some difficulty to understand the problem. When I open the 2 different images with the same class I am not seeing any similarity. For example, images with class 0 002679c2-bbb6-11e8-b2ba-ac1f6b6435d0.png and 001bcdd2-bbb2-11e8-b2ba-ac1f6b6435d0.png doesn't show any similarity. Could it be because of different morpholgies? If yes, then should it be 2 different classes?
\n\nPlease don't mind my ignorance. I don't have any background in the bio-tech.
", "messages": [ { "id": "403942", "postDate": "10/15/2018 00:25:02", "content": "Hello,
\n\nI have some difficulty to understand the problem. When I open the 2 different images with the same class I am not seeing any similarity. For example, images with class 0 002679c2-bbb6-11e8-b2ba-ac1f6b6435d0.png and 001bcdd2-bbb2-11e8-b2ba-ac1f6b6435d0.png doesn't show any similarity. Could it be because of different morpholgies? If yes, then should it be 2 different classes?
\n\nPlease don't mind my ignorance. I don't have any background in the bio-tech.
", "rawMarkdown": "Hello,\n\nI have some difficulty to understand the problem. When I open the 2 different images with the same class I am not seeing any similarity. For example, images with class 0 002679c2-bbb6-11e8-b2ba-ac1f6b6435d0.png and 001bcdd2-bbb2-11e8-b2ba-ac1f6b6435d0.png doesn't show any similarity. Could it be because of different morpholgies? If yes, then should it be 2 different classes?\n\nPlease don't mind my ignorance. I don't have any background in the bio-tech.", "votes": null }, { "id": "404127", "postDate": "10/15/2018 09:35:46", "content": "Hi Shalin, both of these images show a similar pattern for the green channel. Label \"0\" is the nucleoplasm, and if you display green and blue channels side by side for each of these two images, you will see the green channel pattern corresponds to the one seen on the blue channel. It's different (for example, one picture has cells with several nuclei, or possibly nuclei with several lobes, and both green channel show some labeling of the cytosol), but it's similar.
", "rawMarkdown": "Hi Shalin, both of these images show a similar pattern for the green channel. Label \"0\" is the nucleoplasm, and if you display green and blue channels side by side for each of these two images, you will see the green channel pattern corresponds to the one seen on the blue channel. It's different (for example, one picture has cells with several nuclei, or possibly nuclei with several lobes, and both green channel show some labeling of the cytosol), but it's similar.", "votes": null }, { "id": "404197", "postDate": "10/15/2018 12:26:08", "content": "Hello Shalin!
\n\nLike Jonathan wrote, we are looking for similar patterns in the green channel. No cell will express the pattern the exact same as another one but the proteins of interest should locate to the same subcellular structures and thanks to that have a similar pattern in the green channel. The other 3 channels help us understand what we are looking at. For example, if the red channel (the microtubule marker) overlaps with the green we can see that the protein of interest is locating to the microtubules.
\n\nIf it is still unclear, don't hesitate to ask! And if you think of a way that we can improve the challenge explanation to make the information more clear that would also be welcome!
", "rawMarkdown": "Hello Shalin!\n\nLike Jonathan wrote, we are looking for similar patterns in the green channel. No cell will express the pattern the exact same as another one but the proteins of interest should locate to the same subcellular structures and thanks to that have a similar pattern in the green channel. The other 3 channels help us understand what we are looking at. For example, if the red channel (the microtubule marker) overlaps with the green we can see that the protein of interest is locating to the microtubules.\n\nIf it is still unclear, don't hesitate to ask! And if you think of a way that we can improve the challenge explanation to make the information more clear that would also be welcome!", "votes": null }, { "id": "412482", "postDate": "10/30/2018 09:13:35", "content": "Hello Jonathan,
\n\nI am very new to deep learning and using this to learn. I was struggling to understand which filter to use for training and test. As I understand from the competition description, are we only training the green filter images for training and testing? and the rest of the filters for more human understanding?
", "rawMarkdown": "Hello Jonathan,\n\nI am very new to deep learning and using this to learn. I was struggling to understand which filter to use for training and test. As I understand from the competition description, are we only training the green filter images for training and testing? and the rest of the filters for more human understanding?", "votes": null }, { "id": "413208", "postDate": "10/31/2018 13:58:14", "content": "Hello,\nI am very new as well, you can use all the filters for training and testing. Most people seem to use all filters so you need to use tricks to use four filters while many algorithms work with three images (because many image color schemes rely on three values, like RGB). For example, you can add two filters and treat this as one channel.
", "rawMarkdown": "Hello,\nI am very new as well, you can use all the filters for training and testing. Most people seem to use all filters so you need to use tricks to use four filters while many algorithms work with three images (because many image color schemes rely on three values, like RGB). For example, you can add two filters and treat this as one channel.", "votes": null }, { "id": "414634", "postDate": "11/03/2018 08:30:20", "content": "Thanks Jonathan. I finally understood the problem. :)
", "rawMarkdown": "Thanks Jonathan. I finally understood the problem. :)", "votes": null }, { "id": "424959", "postDate": "11/20/2018 23:50:38", "content": "Hello, \nI've read over the description and have downloaded the data. I'm wondering - is there any sort of label that shows the cell type each image is showing (liver, brain, etc)? Assuming patterns can be found among the green channels of two different images, how would one be able to tell which two cell types the protein reoccurs in?
", "rawMarkdown": "Hello, \nI've read over the description and have downloaded the data. I'm wondering - is there any sort of label that shows the cell type each image is showing (liver, brain, etc)? Assuming patterns can be found among the green channels of two different images, how would one be able to tell which two cell types the protein reoccurs in?", "votes": null }, { "id": "424963", "postDate": "11/21/2018 00:01:54", "content": "\n\n\nAssuming patterns can be found among the green channels of two different images, how would one be able to tell which two cell types the protein reoccurs in?
\n
@andstun These images are of over-expressed proteins, meaning that they will be found in that particular cell type by forcing a cell to make them regardless of whether that happens under normal conditions. Don't think that any of these cells are from liver or brain (no \"general\" cell types are), though they probably wouldn't tell us anyway because that would likely diminish the method's generalization. I can recognize with certainty that some images are of HeLa cells (human cervical cancer), and I am pretty confident that some Cos cells are there as well (those are derived from monkey kidney).
", "rawMarkdown": "> Assuming patterns can be found among the green channels of two different images, how would one be able to tell which two cell types the protein reoccurs in?\n\n@andstun These images are of over-expressed proteins, meaning that they will be found in that particular cell type by forcing a cell to make them regardless of whether that happens under normal conditions. Don't think that any of these cells are from liver or brain (no \"general\" cell types are), though they probably wouldn't tell us anyway because that would likely diminish the method's generalization. I can recognize with certainty that some images are of HeLa cells (human cervical cancer), and I am pretty confident that some Cos cells are there as well (those are derived from monkey kidney).", "votes": null }, { "id": "425021", "postDate": "11/21/2018 02:36:42", "content": "Thank you for the explanation! So it doesn't matter what kind of proteins the model is trained on? In other words, a high-accuracy model in this competition could then be used by the Human Protein Atlas team to identify similar patterns within the human body?
", "rawMarkdown": "Thank you for the explanation! So it doesn't matter what kind of proteins the model is trained on? In other words, a high-accuracy model in this competition could then be used by the Human Protein Atlas team to identify similar patterns within the human body?", "votes": null } ], "comments": [ { "id": 404127, "author_name": "jschnab", "author_url": "", "post_date": "10/15/2018 09:35:46", "content": "Hi Shalin, both of these images show a similar pattern for the green channel. Label \"0\" is the nucleoplasm, and if you display green and blue channels side by side for each of these two images, you will see the green channel pattern corresponds to the one seen on the blue channel. It's different (for example, one picture has cells with several nuclei, or possibly nuclei with several lobes, and both green channel show some labeling of the cytosol), but it's similar.
", "votes": null, "replies": [ { "id": 412482, "author_name": "merrilmathew", "author_url": "", "post_date": "10/30/2018 09:13:35", "content": "Hello Jonathan,
\n\nI am very new to deep learning and using this to learn. I was struggling to understand which filter to use for training and test. As I understand from the competition description, are we only training the green filter images for training and testing? and the rest of the filters for more human understanding?
", "votes": null, "replies": [] }, { "id": 413208, "author_name": "jschnab", "author_url": "", "post_date": "10/31/2018 13:58:14", "content": "Hello,\nI am very new as well, you can use all the filters for training and testing. Most people seem to use all filters so you need to use tricks to use four filters while many algorithms work with three images (because many image color schemes rely on three values, like RGB). For example, you can add two filters and treat this as one channel.
", "votes": null, "replies": [] }, { "id": 414634, "author_name": "merrilmathew", "author_url": "", "post_date": "11/03/2018 08:30:20", "content": "Thanks Jonathan. I finally understood the problem. :)
", "votes": null, "replies": [] } ] }, { "id": 404197, "author_name": "cwinsnes", "author_url": "", "post_date": "10/15/2018 12:26:08", "content": "Hello Shalin!
\n\nLike Jonathan wrote, we are looking for similar patterns in the green channel. No cell will express the pattern the exact same as another one but the proteins of interest should locate to the same subcellular structures and thanks to that have a similar pattern in the green channel. The other 3 channels help us understand what we are looking at. For example, if the red channel (the microtubule marker) overlaps with the green we can see that the protein of interest is locating to the microtubules.
\n\nIf it is still unclear, don't hesitate to ask! And if you think of a way that we can improve the challenge explanation to make the information more clear that would also be welcome!
", "votes": null, "replies": [ { "id": 424959, "author_name": "andstun", "author_url": "", "post_date": "11/20/2018 23:50:38", "content": "Hello, \nI've read over the description and have downloaded the data. I'm wondering - is there any sort of label that shows the cell type each image is showing (liver, brain, etc)? Assuming patterns can be found among the green channels of two different images, how would one be able to tell which two cell types the protein reoccurs in?
", "votes": null, "replies": [] }, { "id": 424963, "author_name": "tilii7", "author_url": "", "post_date": "11/21/2018 00:01:54", "content": "\n\n\nAssuming patterns can be found among the green channels of two different images, how would one be able to tell which two cell types the protein reoccurs in?
\n
@andstun These images are of over-expressed proteins, meaning that they will be found in that particular cell type by forcing a cell to make them regardless of whether that happens under normal conditions. Don't think that any of these cells are from liver or brain (no \"general\" cell types are), though they probably wouldn't tell us anyway because that would likely diminish the method's generalization. I can recognize with certainty that some images are of HeLa cells (human cervical cancer), and I am pretty confident that some Cos cells are there as well (those are derived from monkey kidney).
", "votes": null, "replies": [] }, { "id": 425021, "author_name": "andstun", "author_url": "", "post_date": "11/21/2018 02:36:42", "content": "Thank you for the explanation! So it doesn't matter what kind of proteins the model is trained on? In other words, a high-accuracy model in this competition could then be used by the Human Protein Atlas team to identify similar patterns within the human body?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "403942": "Hello,\n\nI have some difficulty to understand the problem. When I open the 2 different images with the same class I am not seeing any similarity. For example, images with class 0 002679c2-bbb6-11e8-b2ba-ac1f6b6435d0.png and 001bcdd2-bbb2-11e8-b2ba-ac1f6b6435d0.png doesn't show any similarity. Could it be because of different morpholgies? If yes, then should it be 2 different classes?\n\nPlease don't mind my ignorance. I don't have any background in the bio-tech.", "404127": "Hi Shalin, both of these images show a similar pattern for the green channel. Label \"0\" is the nucleoplasm, and if you display green and blue channels side by side for each of these two images, you will see the green channel pattern corresponds to the one seen on the blue channel. It's different (for example, one picture has cells with several nuclei, or possibly nuclei with several lobes, and both green channel show some labeling of the cytosol), but it's similar.", "404197": "Hello Shalin!\n\nLike Jonathan wrote, we are looking for similar patterns in the green channel. No cell will express the pattern the exact same as another one but the proteins of interest should locate to the same subcellular structures and thanks to that have a similar pattern in the green channel. The other 3 channels help us understand what we are looking at. For example, if the red channel (the microtubule marker) overlaps with the green we can see that the protein of interest is locating to the microtubules.\n\nIf it is still unclear, don't hesitate to ask! And if you think of a way that we can improve the challenge explanation to make the information more clear that would also be welcome!", "412482": "Hello Jonathan,\n\nI am very new to deep learning and using this to learn. I was struggling to understand which filter to use for training and test. As I understand from the competition description, are we only training the green filter images for training and testing? and the rest of the filters for more human understanding?", "413208": "Hello,\nI am very new as well, you can use all the filters for training and testing. Most people seem to use all filters so you need to use tricks to use four filters while many algorithms work with three images (because many image color schemes rely on three values, like RGB). For example, you can add two filters and treat this as one channel.", "414634": "Thanks Jonathan. I finally understood the problem. :)", "424959": "Hello, \nI've read over the description and have downloaded the data. I'm wondering - is there any sort of label that shows the cell type each image is showing (liver, brain, etc)? Assuming patterns can be found among the green channels of two different images, how would one be able to tell which two cell types the protein reoccurs in?", "424963": "> Assuming patterns can be found among the green channels of two different images, how would one be able to tell which two cell types the protein reoccurs in?\n\n@andstun These images are of over-expressed proteins, meaning that they will be found in that particular cell type by forcing a cell to make them regardless of whether that happens under normal conditions. Don't think that any of these cells are from liver or brain (no \"general\" cell types are), though they probably wouldn't tell us anyway because that would likely diminish the method's generalization. I can recognize with certainty that some images are of HeLa cells (human cervical cancer), and I am pretty confident that some Cos cells are there as well (those are derived from monkey kidney).", "425021": "Thank you for the explanation! So it doesn't matter what kind of proteins the model is trained on? In other words, a high-accuracy model in this competition could then be used by the Human Protein Atlas team to identify similar patterns within the human body?" }, "source": "meta" }