\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F16720711%2Fb78d6b15da3ba104c41c14d23ae09372%2Fimage%20(4).png?generation=1694691834222757&alt=media\")
Dear Kagglers,
\nOn behalf of Open Problem in Single-Cell Analysis, I would like to officially welcome you to the “Open Problems - Single-Cell Perturbations” competition, part of the NeurIPS 2023 Competition Track!
\nThis competition is about predicting how different cell types respond to different drug perturbations. In the field of biomedicine, single-cell technologies have led to an explosion of new findings about the diversity of the 37 trillion cells in the human body and how these cell types vary between health and disease. Efficiently designing new medicines requires being able to model how different kinds of cells respond to different kinds of small molecules.
\nFor this competition, Open Problems partnered for the third year in a row with Cellarity, a cell-centric drug creation company, to generate a first-of-its-kind benchmarking dataset designed to drive advances in algorithms that can capture drivers of changes in cell state over time. This year, we measured peripheral blood mononuclear cells (PBMCs) from 3 healthy human donors responding to 144 different small molecules plus 2 positive controls. These PBMCs are organized into 5 different cell types, each of which respond uniquely to each perturbation. Your challenge will be to learn the patterns behind cell-type specific responses on 4 of the cell types, and predict into two held-out cell types, where you'll only have access to 10% of the compounds.
\nIn the Data section, you’ll find descriptions of the datasets and plenty of metadata on the compounds. Please feel free to ask us any questions!
\nIn our NeurIPS competition last year, we saw over 1,600 competitors come together to work on predicting between modalities. We hope to see such broad participation again.
\nOne new component of the competition this year is the introduction of a Judges Award Track with a $50,000 prize pool for the top 5 submission. The goal is to incentivize scientific contributions that advance our understanding of this problem in addition to top performing models. See full details on the Overview page.
\nThis competition was a collaborative effort between scientists at Cellarity and Helmholtz Munich with sponsorship from Cellarity and the Chan Zuckerberg Initiative's Single-Cell Biology Program.
\nIf you’re interested to learn more about Open Problems in Single-Cell Analysis, please visit our homepage, https://openproblems.bio, where you can sign up for our mailing list.
\nBest of luck! We can’t wait to see what you build.
\nDaniel Burkhardt and the rest of the Core Team at Open Problems
P.S. We've also secured support from our friends at Saturn Cloud to provide upgraded free compute instances with GPU for participants in the competition. Stay tuned for more details 👀
", "messages": [ { "id": "2426899", "postDate": "09/06/2023 22:41:41", "content": "Dear Kagglers,
\nOn behalf of Open Problem in Single-Cell Analysis, I would like to officially welcome you to the “Open Problems - Single-Cell Perturbations” competition, part of the NeurIPS 2023 Competition Track!
\nThis competition is about predicting how different cell types respond to different drug perturbations. In the field of biomedicine, single-cell technologies have led to an explosion of new findings about the diversity of the 37 trillion cells in the human body and how these cell types vary between health and disease. Efficiently designing new medicines requires being able to model how different kinds of cells respond to different kinds of small molecules.
\nFor this competition, Open Problems partnered for the third year in a row with Cellarity, a cell-centric drug creation company, to generate a first-of-its-kind benchmarking dataset designed to drive advances in algorithms that can capture drivers of changes in cell state over time. This year, we measured peripheral blood mononuclear cells (PBMCs) from 3 healthy human donors responding to 144 different small molecules plus 2 positive controls. These PBMCs are organized into 5 different cell types, each of which respond uniquely to each perturbation. Your challenge will be to learn the patterns behind cell-type specific responses on 4 of the cell types, and predict into two held-out cell types, where you'll only have access to 10% of the compounds.
\nIn the Data section, you’ll find descriptions of the datasets and plenty of metadata on the compounds. Please feel free to ask us any questions!
\nIn our NeurIPS competition last year, we saw over 1,600 competitors come together to work on predicting between modalities. We hope to see such broad participation again.
\nOne new component of the competition this year is the introduction of a Judges Award Track with a $50,000 prize pool for the top 5 submission. The goal is to incentivize scientific contributions that advance our understanding of this problem in addition to top performing models. See full details on the Overview page.
\nThis competition was a collaborative effort between scientists at Cellarity and Helmholtz Munich with sponsorship from Cellarity and the Chan Zuckerberg Initiative's Single-Cell Biology Program.
\nIf you’re interested to learn more about Open Problems in Single-Cell Analysis, please visit our homepage, https://openproblems.bio, where you can sign up for our mailing list.
\nBest of luck! We can’t wait to see what you build.
\nDaniel Burkhardt and the rest of the Core Team at Open Problems
P.S. We've also secured support from our friends at Saturn Cloud to provide upgraded free compute instances with GPU for participants in the competition. Stay tuned for more details 👀
", "rawMarkdown": "Dear Kagglers,\n\nOn behalf of Open Problem in Single-Cell Analysis, I would like to officially welcome you to the “Open Problems - Single-Cell Perturbations” competition, part of the NeurIPS 2023 Competition Track!\n\nThis competition is about predicting how different cell types respond to different drug perturbations. In the field of biomedicine, single-cell technologies have led to an explosion of new findings about the diversity of the 37 trillion cells in the human body and how these cell types vary between health and disease. Efficiently designing new medicines requires being able to model how different kinds of cells respond to different kinds of small molecules.\n\nFor this competition, Open Problems partnered for the third year in a row with Cellarity, a cell-centric drug creation company, to generate a first-of-its-kind benchmarking dataset designed to drive advances in algorithms that can capture drivers of changes in cell state over time. This year, we measured peripheral blood mononuclear cells (PBMCs) from 3 healthy human donors responding to 144 different small molecules plus 2 positive controls. These PBMCs are organized into 5 different cell types, each of which respond uniquely to each perturbation. Your challenge will be to learn the patterns behind cell-type specific responses on 4 of the cell types, and predict into two held-out cell types, where you'll only have access to 10% of the compounds.\n\nIn the Data section, you’ll find descriptions of the datasets and plenty of metadata on the compounds. Please feel free to ask us any questions!\n\nIn [our NeurIPS competition last year](https://www.kaggle.com/competitions/open-problems-multimodal/overview), we saw over 1,600 competitors come together to work on predicting between modalities. We hope to see such broad participation again.\n\nOne new component of the competition this year is the introduction of a Judges Award Track with a $50,000 prize pool for the top 5 submission. The goal is to incentivize scientific contributions that advance our understanding of this problem in addition to top performing models. See full details on the Overview page.\n\nThis competition was a collaborative effort between scientists at Cellarity and Helmholtz Munich with sponsorship from Cellarity and the Chan Zuckerberg Initiative's Single-Cell Biology Program.\n\nIf you’re interested to learn more about Open Problems in Single-Cell Analysis, please visit our homepage, https://openproblems.bio, where you can sign up for our mailing list.\n\nBest of luck! We can’t wait to see what you build.\nDaniel Burkhardt and the rest of the Core Team at Open Problems\n\nP.S. We've also secured support from our friends at Saturn Cloud to provide upgraded free compute instances with GPU for participants in the competition. Stay tuned for more details 👀", "votes": null }, { "id": "2438619", "postDate": "09/14/2023 11:44:45", "content": "Hi, the output should be the signed log-fold change. May be there is a typo here
\n
I am thrilled to be part of the 'Open Problems - Single-Cell Perturbations' competition at NeurIPS 2023! The opportunity to delve into the intricacies of single-cell analysis and drug perturbations is truly exciting. As I look forward to this challenging journey, I am also keen on finding like-minded individuals who share a passion for this field and would like to form a team. If anyone is interested in teaming up to tackle this competition together, please don't hesitate to reach out. Looking forward to working together to make a significant contribution to this competition!
\nThank you to Daniel Burkhardt and the entire Core Team at Open Problems for organizing this remarkable event, and for the generous support from Saturn Cloud. I can't wait to see the innovative solutions and discoveries that will emerge from this competition.
\nWarm regards,
\nJoe Sopko
We find that signed -log10(p-values) are a more robust measure of the degree to which a gene is impacted by a perturbation than the log fold change. LFC is more sensitive to the baseline expression of a genes, and can be artificially inflated for lowly expressed genes.
So just to be abundantly clear, the training data also includes the signed -log(10) p-values for the differential expression of each gene, not the actual differential expression fold change correct?
", "rawMarkdown": "So just to be abundantly clear, the training data also includes the signed -log(10) p-values for the differential expression of each gene, not the actual differential expression fold change correct?", "votes": null }, { "id": "2447569", "postDate": "09/20/2023 06:45:59", "content": "Great competition! The dataset is intriguing, and the challenge pushes us to think innovatively. Thanks to the organizers for this opportunity. Looking forward to the results
", "rawMarkdown": "Great competition! The dataset is intriguing, and the challenge pushes us to think innovatively. Thanks to the organizers for this opportunity. Looking forward to the results", "votes": null }, { "id": "2460572", "postDate": "09/28/2023 22:27:15", "content": "Hello!
\nThank you for all the organisers for setup up the competition for us. It seems extremely interesting.
\nIs there any particular reason, why the team sizes are limited to 5 people?
\nIt would be great if this limit could be slightly lifted (e.g 7 people).
Hello, i am interested in teaming up, plz let me know how to coordinate
", "rawMarkdown": "Hello, i am interested in teaming up, plz let me know how to coordinate", "votes": null }, { "id": "2510397", "postDate": "11/03/2023 00:42:23", "content": "Interesting.
", "rawMarkdown": "Interesting.", "votes": null } ], "comments": [ { "id": 2438619, "author_name": "reemalsulamikaust", "author_url": "", "post_date": "09/14/2023 11:44:45", "content": "Hi, the output should be the signed log-fold change. May be there is a typo here
\n
We find that signed -log10(p-values) are a more robust measure of the degree to which a gene is impacted by a perturbation than the log fold change. LFC is more sensitive to the baseline expression of a genes, and can be artificially inflated for lowly expressed genes.
So just to be abundantly clear, the training data also includes the signed -log(10) p-values for the differential expression of each gene, not the actual differential expression fold change correct?
", "votes": null, "replies": [] } ] } ] }, { "id": 2440872, "author_name": "joesopko", "author_url": "", "post_date": "09/15/2023 19:42:12", "content": "I am thrilled to be part of the 'Open Problems - Single-Cell Perturbations' competition at NeurIPS 2023! The opportunity to delve into the intricacies of single-cell analysis and drug perturbations is truly exciting. As I look forward to this challenging journey, I am also keen on finding like-minded individuals who share a passion for this field and would like to form a team. If anyone is interested in teaming up to tackle this competition together, please don't hesitate to reach out. Looking forward to working together to make a significant contribution to this competition!
\nThank you to Daniel Burkhardt and the entire Core Team at Open Problems for organizing this remarkable event, and for the generous support from Saturn Cloud. I can't wait to see the innovative solutions and discoveries that will emerge from this competition.
\nWarm regards,
\nJoe Sopko
Hello, i am interested in teaming up, plz let me know how to coordinate
", "votes": null, "replies": [] } ] }, { "id": 2447569, "author_name": "wsfwsf", "author_url": "", "post_date": "09/20/2023 06:45:59", "content": "Great competition! The dataset is intriguing, and the challenge pushes us to think innovatively. Thanks to the organizers for this opportunity. Looking forward to the results
", "votes": null, "replies": [] }, { "id": 2460572, "author_name": "blazej0", "author_url": "", "post_date": "09/28/2023 22:27:15", "content": "Hello!
\nThank you for all the organisers for setup up the competition for us. It seems extremely interesting.
\nIs there any particular reason, why the team sizes are limited to 5 people?
\nIt would be great if this limit could be slightly lifted (e.g 7 people).
Interesting.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2426899": "Dear Kagglers,\n\nOn behalf of Open Problem in Single-Cell Analysis, I would like to officially welcome you to the “Open Problems - Single-Cell Perturbations” competition, part of the NeurIPS 2023 Competition Track!\n\nThis competition is about predicting how different cell types respond to different drug perturbations. In the field of biomedicine, single-cell technologies have led to an explosion of new findings about the diversity of the 37 trillion cells in the human body and how these cell types vary between health and disease. Efficiently designing new medicines requires being able to model how different kinds of cells respond to different kinds of small molecules.\n\nFor this competition, Open Problems partnered for the third year in a row with Cellarity, a cell-centric drug creation company, to generate a first-of-its-kind benchmarking dataset designed to drive advances in algorithms that can capture drivers of changes in cell state over time. This year, we measured peripheral blood mononuclear cells (PBMCs) from 3 healthy human donors responding to 144 different small molecules plus 2 positive controls. These PBMCs are organized into 5 different cell types, each of which respond uniquely to each perturbation. Your challenge will be to learn the patterns behind cell-type specific responses on 4 of the cell types, and predict into two held-out cell types, where you'll only have access to 10% of the compounds.\n\nIn the Data section, you’ll find descriptions of the datasets and plenty of metadata on the compounds. Please feel free to ask us any questions!\n\nIn [our NeurIPS competition last year](https://www.kaggle.com/competitions/open-problems-multimodal/overview), we saw over 1,600 competitors come together to work on predicting between modalities. We hope to see such broad participation again.\n\nOne new component of the competition this year is the introduction of a Judges Award Track with a $50,000 prize pool for the top 5 submission. The goal is to incentivize scientific contributions that advance our understanding of this problem in addition to top performing models. See full details on the Overview page.\n\nThis competition was a collaborative effort between scientists at Cellarity and Helmholtz Munich with sponsorship from Cellarity and the Chan Zuckerberg Initiative's Single-Cell Biology Program.\n\nIf you’re interested to learn more about Open Problems in Single-Cell Analysis, please visit our homepage, https://openproblems.bio, where you can sign up for our mailing list.\n\nBest of luck! We can’t wait to see what you build.\nDaniel Burkhardt and the rest of the Core Team at Open Problems\n\nP.S. We've also secured support from our friends at Saturn Cloud to provide upgraded free compute instances with GPU for participants in the competition. Stay tuned for more details 👀", "2438619": "Hi, the output should be the signed log-fold change. May be there is a typo here\n.png?generation=1694691834222757&alt=media)", "2440872": "I am thrilled to be part of the 'Open Problems - Single-Cell Perturbations' competition at NeurIPS 2023! The opportunity to delve into the intricacies of single-cell analysis and drug perturbations is truly exciting. As I look forward to this challenging journey, I am also keen on finding like-minded individuals who share a passion for this field and would like to form a team. If anyone is interested in teaming up to tackle this competition together, please don't hesitate to reach out. Looking forward to working together to make a significant contribution to this competition!\n\nThank you to Daniel Burkhardt and the entire Core Team at Open Problems for organizing this remarkable event, and for the generous support from Saturn Cloud. I can't wait to see the innovative solutions and discoveries that will emerge from this competition.\n\nWarm regards,\nJoe Sopko", "2440937": "We find that signed `-log10(p-values)` are a more robust measure of the degree to which a gene is impacted by a perturbation than the log fold change. LFC is more sensitive to the baseline expression of a genes, and can be artificially inflated for lowly expressed genes.", "2443752": "So just to be abundantly clear, the training data also includes the signed -log(10) p-values for the differential expression of each gene, not the actual differential expression fold change correct?", "2447569": "Great competition! The dataset is intriguing, and the challenge pushes us to think innovatively. Thanks to the organizers for this opportunity. Looking forward to the results", "2460572": "Hello!\n\nThank you for all the organisers for setup up the competition for us. It seems extremely interesting.\n\nIs there any particular reason, why the team sizes are limited to 5 people?\nIt would be great if this limit could be slightly lifted (e.g 7 people).", "2480490": "Hello, i am interested in teaming up, plz let me know how to coordinate", "2510397": "Interesting." }, "source": "meta" }