Welcome to the Recursion Cellular Image Classification Challenge! We're excited to be providing free TPU time to participants to enable blazing fast model training. See the Overview:Resources page for details and the signup form. TPUs do impose some additional standardization constraints in order to achieve their high performance, so I would encourage you to take a look at the sample TPU training code Recusion Pharmaceuticals has provided.
\n\nLastly, we've laid out the core elements of this image classification problem on the data page, but if you're curious about the problem's background or just want to know what the heck an siRNA is you should also take a look at the rxrx.ai. As a lapsed biologist I may be biased, but I think siRNA is a really cool biological tool and that it's worth reading up on how Recursion is using it to conduct research at scale.
", "messages": [ { "id": 563049, "postDate": "2019-06-27T19:49:08.100Z", "content": "Welcome to the Recursion Cellular Image Classification Challenge! We're excited to be providing free TPU time to participants to enable blazing fast model training. See the Overview:Resources page for details and the signup form. TPUs do impose some additional standardization constraints in order to achieve their high performance, so I would encourage you to take a look at the sample TPU training code Recusion Pharmaceuticals has provided.
\n\nLastly, we've laid out the core elements of this image classification problem on the data page, but if you're curious about the problem's background or just want to know what the heck an siRNA is you should also take a look at the rxrx.ai. As a lapsed biologist I may be biased, but I think siRNA is a really cool biological tool and that it's worth reading up on how Recursion is using it to conduct research at scale.
", "rawMarkdown": "Welcome to the Recursion Cellular Image Classification Challenge! We're excited to be providing free [TPU](https://cloud.google.com/tpu/) time to participants to enable blazing fast model training. See the [Overview:Resources page](https://www.kaggle.com/c/recursion-cellular-image-classification/overview/resources) for details and the signup form. TPUs do impose some additional standardization constraints in order to achieve their high performance, so I would encourage you to take a look at the [sample TPU training code](https://github.com/recursionpharma/rxrx1-utils) Recusion Pharmaceuticals has provided.\n\nLastly, we've laid out the core elements of this image classification problem on [the data page](https://www.kaggle.com/c/recursion-cellular-image-classification/data), but if you're curious about the problem's background or just want to know what the heck an siRNA is you should also take a look at the [rxrx.ai](https://www.rxrx.ai/). As a lapsed biologist I may be biased, but I think siRNA is a really cool biological tool and that it's worth reading up on how Recursion is using it to conduct research at scale.\n", "votes": 9 }, { "id": 563977, "postDate": "2019-06-28T20:56:05.097Z", "content": "Thanks! The link to rxrx.ai is wrong, it should point to https://www.rxrx.ai/.
", "rawMarkdown": "Thanks! The link to rxrx.ai is wrong, it should point to https://www.rxrx.ai/.", "votes": 1, "replies": [ { "id": 564051, "postDate": "2019-06-28T23:43:22.093Z", "content": "Thank you, fixed.
", "rawMarkdown": "Thank you, fixed.", "votes": 1 } ] }, { "id": 564451, "postDate": "2019-06-29T12:47:53.973Z", "content": "What is the point of predicting siRNA for every particular experiment, plate and well? You definitely know what siRNA you have applied to a given well, am I correct?
", "rawMarkdown": "What is the point of predicting siRNA for every particular experiment, plate and well? You definitely know what siRNA you have applied to a given well, am I correct?", "votes": 2, "replies": [ { "id": 564747, "postDate": "2019-06-29T23:15:58.563Z", "content": "I'm not sure I understand the question, but I can confirm that the location of siRNAs within an experiment are randomized with respect to plate and well.
", "rawMarkdown": "I'm not sure I understand the question, but I can confirm that the location of siRNAs within an experiment are randomized with respect to plate and well." }, { "id": 564751, "postDate": "2019-06-29T23:39:50.737Z", "content": "Hi Berton! I'm just trying to understand how things work for you. Let's suppose that you've got a model that perfectly predicts siRNA by the given image (99.99% accuracy). How would you use it?
\n\nI thought the point is to save time on experiments by using an ML predictor model. But you don't ask us to predict an outcome of experiments. What am I missing here? Thanks.
", "rawMarkdown": "Hi Berton! I'm just trying to understand how things work for you. Let's suppose that you've got a model that perfectly predicts siRNA by the given image (99.99% accuracy). How would you use it?\n\nI thought the point is to save time on experiments by using an ML predictor model. But you don't ask us to predict an outcome of experiments. What am I missing here? Thanks." }, { "id": 564795, "postDate": "2019-06-30T01:46:22.357Z", "content": "Right, now I understand, good question. This competition and the RxRx1 dataset are not primarily about predicting the outcome of any particular experiment, but for exploring techniques for disentangling biological signal (siRNA) from technical signal (batch and plate effects). We measure this by seeing how well models predict the biological signal in batches held out from the training set.
", "rawMarkdown": "Right, now I understand, good question. This competition and the RxRx1 dataset are not primarily about predicting the outcome of any particular experiment, but for exploring techniques for disentangling biological signal (siRNA) from technical signal (batch and plate effects). We measure this by seeing how well models predict the biological signal in batches held out from the training set.", "votes": 3 } ] }, { "id": 564268, "postDate": "2019-06-29T08:06:55.867Z", "content": "Thank you! Small question, why do we have siRNA value 1138 for negative controls? In the description on https://www.rxrx.ai/ it says that that negative control should have no treatment. Is 1138 a placeholder value, or is this a different kind of treatment? I know that this siRNA value is absent in train, still want to confirm.
", "rawMarkdown": "Thank you! Small question, why do we have siRNA value 1138 for negative controls? In the description on https://www.rxrx.ai/ it says that that negative control should have no treatment. Is 1138 a placeholder value, or is this a different kind of treatment? I know that this siRNA value is absent in train, still want to confirm.", "votes": 1, "replies": [ { "id": 564746, "postDate": "2019-06-29T23:14:21.933Z", "content": "Yes, 1138 is just a placeholder value for no treatment.
", "rawMarkdown": "Yes, 1138 is just a placeholder value for no treatment.", "votes": 2 } ] }, { "id": 580814, "postDate": "2019-07-20T19:57:59.947Z", "content": "Are the different siRNA's guaranteed to be inducing a distinct visual phenotype
", "rawMarkdown": "Are the different siRNA's guaranteed to be inducing a distinct visual phenotype", "replies": [ { "id": 600742, "postDate": "2019-08-16T14:08:04.770Z", "content": "", "rawMarkdown": "@sohier @bearnshaw " } ] }, { "id": 580099, "postDate": "2019-07-19T17:12:39.340Z", "content": "@sohier Hi, our best score is not shown in public lb and our rank is also incorrect. 😂 \nEdit: Thank you. It is fixed.
", "rawMarkdown": "@sohier Hi, our best score is not shown in public lb and our rank is also incorrect. 😂 \nEdit: Thank you. It is fixed." }, { "id": 577759, "postDate": "2019-07-17T03:39:46.220Z", "content": "Hello, this is my first time participating in a competition and I have a stupid question. Can we modify / re-use the code shared on GITHUB shared under resources?
", "rawMarkdown": "Hello, this is my first time participating in a competition and I have a stupid question. Can we modify / re-use the code shared on GITHUB shared under resources?", "replies": [ { "id": 589426, "postDate": "2019-07-31T21:49:35.183Z", "content": "Sure.
", "rawMarkdown": "Sure." } ] }, { "id": 577374, "postDate": "2019-07-16T16:32:30.340Z", "content": "Thank you!
\n\nA suggestion, could you make a \"zarrified\" (https://github.com/recursionpharma/rxrx1-utils/blob/master/rxrx/preprocess/images2zarr.py) version of data for the ease of use with Kaggle Kernels?
\n\nData loading with the current format is quite slow in kernels. In my quick experiment (ResNet50, 224x224 input, batch size 128), the network runs 0.5s/batch, but loading a batch could consume 2 seconds.
", "rawMarkdown": "Thank you!\n\nA suggestion, could you make a \"zarrified\" (https://github.com/recursionpharma/rxrx1-utils/blob/master/rxrx/preprocess/images2zarr.py) version of data for the ease of use with Kaggle Kernels?\n\nData loading with the current format is quite slow in kernels. In my quick experiment (ResNet50, 224x224 input, batch size 128), the network runs 0.5s/batch, but loading a batch could consume 2 seconds." }, { "id": 568287, "postDate": "2019-07-04T16:40:27.600Z", "content": "First competition entry for me so another silly question. On the submissions entry page I can see you are expecting 19897 entries. I make it 44286 in the test set, of which 39794 are treatment. Can you help me lock down where I should be making a prediction?
", "rawMarkdown": "First competition entry for me so another silly question. On the submissions entry page I can see you are expecting 19897 entries. I make it 44286 in the test set, of which 39794 are treatment. Can you help me lock down where I should be making a prediction?", "replies": [ { "id": 568376, "postDate": "2019-07-04T19:52:18.447Z", "content": "There are two images (sites) per row in the submission file (= per well), hence 39794 images divided by 2 becomes 19897.
", "rawMarkdown": "There are two images (sites) per row in the submission file (= per well), hence 39794 images divided by 2 becomes 19897.", "votes": 1 }, { "id": 568602, "postDate": "2019-07-05T07:31:28.730Z", "content": "Thanks very much
", "rawMarkdown": "Thanks very much" } ] }, { "id": 566858, "postDate": "2019-07-02T16:59:51.950Z", "content": "Very likely a very silly question... how do nuclei and nucleoli relate?
\n\nI've had to turn to the internet to better understand the functions and purposes of the parts of the cell that each layer is picking up. I'm reading that a nucleolus (singular nucleoli) is contained within a nuclei. Some sources that I looked at - https://www.quora.com/Whats-the-difference-between-nucleus-nucleolus-and-nucleoli-in-Biology\nhttps://en.wikipedia.org/wiki/Nucleolus
\n\nOn the Fig 6 description -
\n\n\n\n\n'Figure 6: The top-left image is a composite of the 6 channels. It is followed by each of the 6 individual channel faux-colored images of HUVEC cells: nuclei (blue), endoplasmic reticuli (green), actin (red), nucleoli (cyan), mitochondria (magenta), and golgi apparatus (yellow). The overlap in channel content is due in part to the lack of complete spectral separation between fluorescent stains.'
\n
I'm reading that channel 1 picks up the nuclei and that channel 4 picks up the nucleoli. To my eye, it looks like the nucleoli of the image are sat outside of the nuclei, contrary to my current understanding of how they relate to each other.
\n\nPlease can you help me understand their relationship / point me to a good resource for better understanding how these elements relate to each other and how they work within the cell.
", "rawMarkdown": "Very likely a very silly question... how do nuclei and nucleoli relate?\n\nI've had to turn to the internet to better understand the functions and purposes of the parts of the cell that each layer is picking up. I'm reading that a nucleolus (singular nucleoli) is contained within a nuclei. Some sources that I looked at - https://www.quora.com/Whats-the-difference-between-nucleus-nucleolus-and-nucleoli-in-Biology\nhttps://en.wikipedia.org/wiki/Nucleolus\n\nOn the Fig 6 description - \n> 'Figure 6: The top-left image is a composite of the 6 channels. It is followed by each of the 6 individual channel faux-colored images of HUVEC cells: nuclei (blue), endoplasmic reticuli (green), actin (red), nucleoli (cyan), mitochondria (magenta), and golgi apparatus (yellow). The overlap in channel content is due in part to the lack of complete spectral separation between fluorescent stains.' \n\nI'm reading that channel 1 picks up the nuclei and that channel 4 picks up the nucleoli. To my eye, it looks like the nucleoli of the image are sat outside of the nuclei, contrary to my current understanding of how they relate to each other. \n\nPlease can you help me understand their relationship / point me to a good resource for better understanding how these elements relate to each other and how they work within the cell.", "replies": [ { "id": 567098, "postDate": "2019-07-03T04:04:00.107Z", "content": "Nucleoli are most often associated with ribosome biogenesis and are main sites for rRNA synthesis.
\n\nThey are indeed within the nucleus except in cases of pathology.
", "rawMarkdown": "Nucleoli are most often associated with ribosome biogenesis and are main sites for rRNA synthesis. \n\nThey are indeed within the nucleus except in cases of pathology. ", "votes": 1 }, { "id": 567495, "postDate": "2019-07-03T15:38:18.800Z", "content": "Thanks very much for clarifying that for me
", "rawMarkdown": "Thanks very much for clarifying that for me" } ] }, { "id": 565006, "postDate": "2019-06-30T09:39:07.140Z", "content": "Thank you!
\n\nI think I'm not getting the point about of non-controlled wells (no scientific background here!). Because they are no controlled, this means that we cannot assure if they have some siRNA effect or not. We could only consider no siRNA effect in the negative controlled ones?\nHow we could take advantatge of all these images? Considering as a negative results or not?
\n\nThank you in advance!\nDavid
", "rawMarkdown": "Thank you!\n\nI think I'm not getting the point about of non-controlled wells (no scientific background here!). Because they are no controlled, this means that we cannot assure if they have some siRNA effect or not. We could only consider no siRNA effect in the negative controlled ones?\nHow we could take advantatge of all these images? Considering as a negative results or not? \n\nThank you in advance!\nDavid", "replies": [ { "id": 565013, "postDate": "2019-06-30T09:44:58.403Z", "content": "\nPositive and Negative Controls\nIn each experiment, the same 30 siRNAs appear on every plate as positive controls. In addition, there is one well per plate that is left untreated as a negative control. \n\nSo, negative control just means untreated cells, no siRNA.
Thank you Artyom!
\n\nI think that's correct. What I'm wondering is about the no-control wells. Should we consider thay can have batch and plate effects but not siRNA effects?
\n\nUPDATE: Well, I'm reviewing and reading that non-control wells are treated with the 1108 siRNAs. Then, what I'm not getting is the concept of control and non-control wells.
\n\nThank you!\nDavid
", "rawMarkdown": "Thank you Artyom!\n\nI think that's correct. What I'm wondering is about the no-control wells. Should we consider thay can have batch and plate effects but not siRNA effects? \n\nUPDATE: Well, I'm reviewing and reading that non-control wells are treated with the 1108 siRNAs. Then, what I'm not getting is the concept of control and non-control wells. \n\nThank you!\nDavid" }, { "id": 567059, "postDate": "2019-07-03T02:25:31.077Z", "content": "> I think that's correct. What I'm wondering is about the no-control wells. Should we consider thay can have batch and plate effects but not siRNA effects?
\n\nIf I am understanding correctly, depending on the well conditions there can be batch and plate effects on them respectively.
", "rawMarkdown": "> I think that's correct. What I'm wondering is about the no-control wells. Should we consider thay can have batch and plate effects but not siRNA effects?\n\nIf I am understanding correctly, depending on the well conditions there can be batch and plate effects on them respectively." }, { "id": 568425, "postDate": "2019-07-04T23:02:41.980Z", "content": "Thank you Pratik!
", "rawMarkdown": "Thank you Pratik!" }, { "id": 572353, "postDate": "2019-07-10T19:59:23.343Z", "rawMarkdown": "", "isDeleted": true }, { "id": 573643, "postDate": "2019-07-12T14:52:02.743Z", "content": "Thank you Thomas!\nI think I finally got the point.
", "rawMarkdown": "Thank you Thomas!\nI think I finally got the point. " } ] }, { "id": 564695, "postDate": "2019-06-29T20:38:38.800Z", "content": "Thanks!
", "rawMarkdown": "Thanks!" }, { "id": 563312, "postDate": "2019-06-28T06:04:48.483Z", "content": "Thanks
", "rawMarkdown": "Thanks" } ], "comments": [ { "id": 563977, "author_name": "Artyom Palvelev", "author_url": "", "post_date": "2019-06-28T20:56:05.097000", "content": "Thanks! The link to rxrx.ai is wrong, it should point to https://www.rxrx.ai/.
", "votes": 1, "replies": [ { "id": 564051, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-06-28T23:43:22.093000", "content": "Thank you, fixed.
", "votes": 1, "replies": [] } ] }, { "id": 564451, "author_name": "Artyom Palvelev", "author_url": "", "post_date": "2019-06-29T12:47:53.973000", "content": "What is the point of predicting siRNA for every particular experiment, plate and well? You definitely know what siRNA you have applied to a given well, am I correct?
", "votes": 2, "replies": [ { "id": 564747, "author_name": "Berton", "author_url": "", "post_date": "2019-06-29T23:15:58.563000", "content": "I'm not sure I understand the question, but I can confirm that the location of siRNAs within an experiment are randomized with respect to plate and well.
", "votes": 0, "replies": [] }, { "id": 564751, "author_name": "Artyom Palvelev", "author_url": "", "post_date": "2019-06-29T23:39:50.737000", "content": "Hi Berton! I'm just trying to understand how things work for you. Let's suppose that you've got a model that perfectly predicts siRNA by the given image (99.99% accuracy). How would you use it?
\n\nI thought the point is to save time on experiments by using an ML predictor model. But you don't ask us to predict an outcome of experiments. What am I missing here? Thanks.
", "votes": 0, "replies": [] }, { "id": 564795, "author_name": "Berton", "author_url": "", "post_date": "2019-06-30T01:46:22.357000", "content": "Right, now I understand, good question. This competition and the RxRx1 dataset are not primarily about predicting the outcome of any particular experiment, but for exploring techniques for disentangling biological signal (siRNA) from technical signal (batch and plate effects). We measure this by seeing how well models predict the biological signal in batches held out from the training set.
", "votes": 3, "replies": [] } ] }, { "id": 564268, "author_name": "Konstantin Lopukhin", "author_url": "", "post_date": "2019-06-29T08:06:55.867000", "content": "Thank you! Small question, why do we have siRNA value 1138 for negative controls? In the description on https://www.rxrx.ai/ it says that that negative control should have no treatment. Is 1138 a placeholder value, or is this a different kind of treatment? I know that this siRNA value is absent in train, still want to confirm.
", "votes": 1, "replies": [ { "id": 564746, "author_name": "Berton", "author_url": "", "post_date": "2019-06-29T23:14:21.933000", "content": "Yes, 1138 is just a placeholder value for no treatment.
", "votes": 2, "replies": [] } ] }, { "id": 580814, "author_name": "Yusuf Roohani", "author_url": "", "post_date": "2019-07-20T19:57:59.947000", "content": "Are the different siRNA's guaranteed to be inducing a distinct visual phenotype
", "votes": 0, "replies": [ { "id": 600742, "author_name": "Yusuf Roohani", "author_url": "", "post_date": "2019-08-16T14:08:04.770000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 580099, "author_name": "Jiwei Liu", "author_url": "", "post_date": "2019-07-19T17:12:39.340000", "content": "@sohier Hi, our best score is not shown in public lb and our rank is also incorrect. 😂 \nEdit: Thank you. It is fixed.
", "votes": 0, "replies": [] }, { "id": 577759, "author_name": "Harsh Patel", "author_url": "", "post_date": "2019-07-17T03:39:46.220000", "content": "Hello, this is my first time participating in a competition and I have a stupid question. Can we modify / re-use the code shared on GITHUB shared under resources?
", "votes": 0, "replies": [ { "id": 589426, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-07-31T21:49:35.183000", "content": "Sure.
", "votes": 0, "replies": [] } ] }, { "id": 577374, "author_name": "tascj", "author_url": "", "post_date": "2019-07-16T16:32:30.340000", "content": "Thank you!
\n\nA suggestion, could you make a \"zarrified\" (https://github.com/recursionpharma/rxrx1-utils/blob/master/rxrx/preprocess/images2zarr.py) version of data for the ease of use with Kaggle Kernels?
\n\nData loading with the current format is quite slow in kernels. In my quick experiment (ResNet50, 224x224 input, batch size 128), the network runs 0.5s/batch, but loading a batch could consume 2 seconds.
", "votes": 0, "replies": [] }, { "id": 568287, "author_name": "Tumble Dyer", "author_url": "", "post_date": "2019-07-04T16:40:27.600000", "content": "First competition entry for me so another silly question. On the submissions entry page I can see you are expecting 19897 entries. I make it 44286 in the test set, of which 39794 are treatment. Can you help me lock down where I should be making a prediction?
", "votes": 0, "replies": [ { "id": 568376, "author_name": "Konstantin Lopukhin", "author_url": "", "post_date": "2019-07-04T19:52:18.447000", "content": "There are two images (sites) per row in the submission file (= per well), hence 39794 images divided by 2 becomes 19897.
", "votes": 1, "replies": [] }, { "id": 568602, "author_name": "Tumble Dyer", "author_url": "", "post_date": "2019-07-05T07:31:28.730000", "content": "Thanks very much
", "votes": 0, "replies": [] } ] }, { "id": 566858, "author_name": "Tumble Dyer", "author_url": "", "post_date": "2019-07-02T16:59:51.950000", "content": "Very likely a very silly question... how do nuclei and nucleoli relate?
\n\nI've had to turn to the internet to better understand the functions and purposes of the parts of the cell that each layer is picking up. I'm reading that a nucleolus (singular nucleoli) is contained within a nuclei. Some sources that I looked at - https://www.quora.com/Whats-the-difference-between-nucleus-nucleolus-and-nucleoli-in-Biology\nhttps://en.wikipedia.org/wiki/Nucleolus
\n\nOn the Fig 6 description -
\n\n\n\n\n'Figure 6: The top-left image is a composite of the 6 channels. It is followed by each of the 6 individual channel faux-colored images of HUVEC cells: nuclei (blue), endoplasmic reticuli (green), actin (red), nucleoli (cyan), mitochondria (magenta), and golgi apparatus (yellow). The overlap in channel content is due in part to the lack of complete spectral separation between fluorescent stains.'
\n
I'm reading that channel 1 picks up the nuclei and that channel 4 picks up the nucleoli. To my eye, it looks like the nucleoli of the image are sat outside of the nuclei, contrary to my current understanding of how they relate to each other.
\n\nPlease can you help me understand their relationship / point me to a good resource for better understanding how these elements relate to each other and how they work within the cell.
", "votes": 0, "replies": [ { "id": 567098, "author_name": "interneuron", "author_url": "", "post_date": "2019-07-03T04:04:00.107000", "content": "Nucleoli are most often associated with ribosome biogenesis and are main sites for rRNA synthesis.
\n\nThey are indeed within the nucleus except in cases of pathology.
", "votes": 1, "replies": [] }, { "id": 567495, "author_name": "Tumble Dyer", "author_url": "", "post_date": "2019-07-03T15:38:18.800000", "content": "Thanks very much for clarifying that for me
", "votes": 0, "replies": [] } ] }, { "id": 565006, "author_name": "David Ibáñez", "author_url": "", "post_date": "2019-06-30T09:39:07.140000", "content": "Thank you!
\n\nI think I'm not getting the point about of non-controlled wells (no scientific background here!). Because they are no controlled, this means that we cannot assure if they have some siRNA effect or not. We could only consider no siRNA effect in the negative controlled ones?\nHow we could take advantatge of all these images? Considering as a negative results or not?
\n\nThank you in advance!\nDavid
", "votes": 0, "replies": [ { "id": 565013, "author_name": "Artyom Palvelev", "author_url": "", "post_date": "2019-06-30T09:44:58.403000", "content": "\nPositive and Negative Controls\nIn each experiment, the same 30 siRNAs appear on every plate as positive controls. In addition, there is one well per plate that is left untreated as a negative control. \n\nSo, negative control just means untreated cells, no siRNA.
Thank you Artyom!
\n\nI think that's correct. What I'm wondering is about the no-control wells. Should we consider thay can have batch and plate effects but not siRNA effects?
\n\nUPDATE: Well, I'm reviewing and reading that non-control wells are treated with the 1108 siRNAs. Then, what I'm not getting is the concept of control and non-control wells.
\n\nThank you!\nDavid
", "votes": 0, "replies": [] }, { "id": 567059, "author_name": "Pratik Gandhi", "author_url": "", "post_date": "2019-07-03T02:25:31.077000", "content": "> I think that's correct. What I'm wondering is about the no-control wells. Should we consider thay can have batch and plate effects but not siRNA effects?
\n\nIf I am understanding correctly, depending on the well conditions there can be batch and plate effects on them respectively.
", "votes": 0, "replies": [] }, { "id": 568425, "author_name": "David Ibáñez", "author_url": "", "post_date": "2019-07-04T23:02:41.980000", "content": "Thank you Pratik!
", "votes": 0, "replies": [] }, { "id": 572353, "author_name": "", "author_url": "", "post_date": "2019-07-10T19:59:23.343000", "content": "", "votes": 0, "replies": [] }, { "id": 573643, "author_name": "David Ibáñez", "author_url": "", "post_date": "2019-07-12T14:52:02.743000", "content": "Thank you Thomas!\nI think I finally got the point.
", "votes": 0, "replies": [] } ] }, { "id": 564695, "author_name": "Abdelghafour Mourchid", "author_url": "", "post_date": "2019-06-29T20:38:38.800000", "content": "Thanks!
", "votes": 0, "replies": [] }, { "id": 563312, "author_name": "shilpa.rpns", "author_url": "", "post_date": "2019-06-28T06:04:48.483000", "content": "Thanks
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "563049": "Welcome to the Recursion Cellular Image Classification Challenge! We're excited to be providing free [TPU](https://cloud.google.com/tpu/) time to participants to enable blazing fast model training. See the [Overview:Resources page](https://www.kaggle.com/c/recursion-cellular-image-classification/overview/resources) for details and the signup form. TPUs do impose some additional standardization constraints in order to achieve their high performance, so I would encourage you to take a look at the [sample TPU training code](https://github.com/recursionpharma/rxrx1-utils) Recusion Pharmaceuticals has provided.\n\nLastly, we've laid out the core elements of this image classification problem on [the data page](https://www.kaggle.com/c/recursion-cellular-image-classification/data), but if you're curious about the problem's background or just want to know what the heck an siRNA is you should also take a look at the [rxrx.ai](https://www.rxrx.ai/). As a lapsed biologist I may be biased, but I think siRNA is a really cool biological tool and that it's worth reading up on how Recursion is using it to conduct research at scale.\n", "563977": "Thanks! The link to rxrx.ai is wrong, it should point to https://www.rxrx.ai/.", "564451": "What is the point of predicting siRNA for every particular experiment, plate and well? You definitely know what siRNA you have applied to a given well, am I correct?", "564268": "Thank you! Small question, why do we have siRNA value 1138 for negative controls? In the description on https://www.rxrx.ai/ it says that that negative control should have no treatment. Is 1138 a placeholder value, or is this a different kind of treatment? I know that this siRNA value is absent in train, still want to confirm.", "580814": "Are the different siRNA's guaranteed to be inducing a distinct visual phenotype", "580099": "@sohier Hi, our best score is not shown in public lb and our rank is also incorrect. 😂 \nEdit: Thank you. It is fixed.", "577759": "Hello, this is my first time participating in a competition and I have a stupid question. Can we modify / re-use the code shared on GITHUB shared under resources?", "577374": "Thank you!\n\nA suggestion, could you make a \"zarrified\" (https://github.com/recursionpharma/rxrx1-utils/blob/master/rxrx/preprocess/images2zarr.py) version of data for the ease of use with Kaggle Kernels?\n\nData loading with the current format is quite slow in kernels. In my quick experiment (ResNet50, 224x224 input, batch size 128), the network runs 0.5s/batch, but loading a batch could consume 2 seconds.", "568287": "First competition entry for me so another silly question. On the submissions entry page I can see you are expecting 19897 entries. I make it 44286 in the test set, of which 39794 are treatment. Can you help me lock down where I should be making a prediction?", "566858": "Very likely a very silly question... how do nuclei and nucleoli relate?\n\nI've had to turn to the internet to better understand the functions and purposes of the parts of the cell that each layer is picking up. I'm reading that a nucleolus (singular nucleoli) is contained within a nuclei. Some sources that I looked at - https://www.quora.com/Whats-the-difference-between-nucleus-nucleolus-and-nucleoli-in-Biology\nhttps://en.wikipedia.org/wiki/Nucleolus\n\nOn the Fig 6 description - \n> 'Figure 6: The top-left image is a composite of the 6 channels. It is followed by each of the 6 individual channel faux-colored images of HUVEC cells: nuclei (blue), endoplasmic reticuli (green), actin (red), nucleoli (cyan), mitochondria (magenta), and golgi apparatus (yellow). The overlap in channel content is due in part to the lack of complete spectral separation between fluorescent stains.' \n\nI'm reading that channel 1 picks up the nuclei and that channel 4 picks up the nucleoli. To my eye, it looks like the nucleoli of the image are sat outside of the nuclei, contrary to my current understanding of how they relate to each other. \n\nPlease can you help me understand their relationship / point me to a good resource for better understanding how these elements relate to each other and how they work within the cell.", "565006": "Thank you!\n\nI think I'm not getting the point about of non-controlled wells (no scientific background here!). Because they are no controlled, this means that we cannot assure if they have some siRNA effect or not. We could only consider no siRNA effect in the negative controlled ones?\nHow we could take advantatge of all these images? Considering as a negative results or not? \n\nThank you in advance!\nDavid", "564695": "Thanks!", "563312": "Thanks" } }