{
  "id": 438868,
  "title": "Single Cell Perturbation. PerturbNet (Deep Generative Model) MichiGAN (NN Architecture).",
  "url": "/competitions/open-problems-single-cell-perturbations/discussion/438868",
  "author_name": "",
  "post_date": "2023-09-12T20:37:41.166985Z",
  "votes": 41,
  "comment_count": 12,
  "views": 0,
  "content": "<h1>PerturbNet, MichiGAN</h1>\n<p>Deep Generative Models for Single-Cell Perturbation Experiments</p>\n<p>Author: Yu, Hengshi (2022)</p>\n<p>\"Recent developments in deep learning have enabled generation of novel and realistic images or sentences from low-dimensional representations. In addition, a revolution in biotechnology has enabled high-throughput measurement of gene expression in thousands to millions of single cells.\"</p>\n<p>\"Several deep generative models have been developed to learn latent representations of cells and generate realistic high-dimensional single-cell data. However, these deep generative models primarily generate data similar to that seen during training and have limited ability to predict gene expression of unseen cell states. In consequence, it constrains the applicability of deep generative models for single-cell data, which usually have a relatively small set of observed conditions. \"</p>\n<p>In Chapter II, the author studied two main classes of deep generative models for single-cell RNA-seq data: variational autoencoders (VAEs) and generative adversarial networks (GANs). He systematically assessed their disentanglement and generation performance and show that VAEs excel at learning cellular representations, while GANs excel at generating realistic single-cell gene expression data.\"</p>\n<p>\"He also developed MichiGAN, a novel neural network architecture that combines the strengths of VAEs and GANs to sample from disentangled representations without sacrificing data generation quality.\"</p>\n<p>\"In Chapter III, he developed PerturbNet, a novel deep generative model to generate single-cell data under unseen drug treatments. Existing approaches attempt to learn drug effects independently of cell state and cannot predict results for unseen drug treatments. To address these limitations, our PerturbNet framework learns mapping from a continuous representation of drug treatment to cellular states. PerturbNet can then generate single-cell data for both observed and unseen drug treatments. He showed that PerturbNet accurately predicts single-cell RNA-seq data resulting from unseen drug treatments. He also fine-tune PerturbNet using cellular properties to improve the continuous representations of drug treatments.\"</p>\n<p>\"In Chapter IV, he extend PerturbNet to learn single-cell responses to genetic perturbations, including pooled CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) genetic inactivations and genetic mutations. Existing approaches attempt to learn genetic perturbation effects independently of cell state and rely on one-hot encodings of genetic perturbations. He developed a GenotypeVAE model and also employ a state-of-the-art protein sequence embedding model to encode genetic perturbations into continuous representations, allowing prediction for both unseen genes and unseen gene combinations.\"</p>\n<p>\"In Chapter V, he extended PerturbNet to design optimal perturbations and attribute perturbation outcomes to specific perturbation features. He considered the translation of a group of cells to a target cell state, and propose two algorithms to design perturbations that achieve this desired target cell state. He showed that the algorithms are effective at designing perturbations that achieve the cell state translation of interest. He also employed model interpretability methods to attribute the effects of chemical or genetic perturbations to specific atoms or gene functional annotations.\"</p>\n<p><a href=\"https://deepblue.lib.umich.edu/handle/2027.42/174555\" target=\"_blank\">https://deepblue.lib.umich.edu/handle/2027.42/174555</a></p>\n<h1>Harmonized Single-Cell Perturbation Data</h1>\n<p>scPerturb: Information Resource for Harmonized Single-Cell Perturbation Data</p>\n<p>Authors: Stefan Peidli, Tessa D. Green, Ciyue Shen, Torsten Gross, Joseph Min, Jake P. Taylor-King, Debora S. Marks, Augustin Luna, Nils Blüthgen, Chris Sander<br>\ndoi: <a href=\"https://doi.org/10.1101/2022.08.20.504663\" target=\"_blank\">https://doi.org/10.1101/2022.08.20.504663</a></p>\n<p>\"Recent biotechnological advances led to growing numbers of single-cell studies, which reveal molecular and phenotypic responses to large numbers of perturbations. However, analysis across diverse datasets is typically hampered by differences in format, naming conventions, data filtering and normalization. In order to facilitate development and benchmarking of computational methods in systems biology, the authors collected a set of 44 publicly available single-cell perturbation-response datasets with molecular readouts, including transcriptomics, proteomics and epigenomics.\"</p>\n<p>\"They applied uniform pre-processing and quality control pipelines and harmonize feature annotations. The resulting information resource enables efficient development and testing of computational analysis methods, and facilitates direct comparison and integration across datasets. Using these datasets, they demonstrated the application of E-distance for quantifying perturbation similarity and strength.\" </p>\n<p><a href=\"https://www.biorxiv.org/content/10.1101/2022.08.20.504663v1.article-metrics\" target=\"_blank\">https://www.biorxiv.org/content/10.1101/2022.08.20.504663v1.article-metrics</a></p>\n<h1>scPerturb: A resource and a python tool for single-cell perturbation data</h1>\n<p>pip install scperturb</p>\n<p>E-distances<br>\nestats = edist(adata, obs_key='perturbation')<br>\n E-distances to a specific group (e.g. 'control')<br>\nestats_control = estats.loc['control']<br>\n E-test for difference to control<br>\ndf = etest(adata, obs_key='perturbation', obsm_key='X_pca', dist='sqeuclidean', control='control', alpha=0.05, runs=100)</p>\n<p><a href=\"https://github.com/sanderlab/scPerturb\" target=\"_blank\">https://github.com/sanderlab/scPerturb</a><br>\n<a href=\"http://projects.sanderlab.org/scperturb/\" target=\"_blank\">http://projects.sanderlab.org/scperturb/</a></p>",
  "messages": [
    {
      "id": "2435272",
      "postDate": "09/12/2023 20:37:41",
      "content": "<h1>PerturbNet, MichiGAN</h1>\n<p>Deep Generative Models for Single-Cell Perturbation Experiments</p>\n<p>Author: Yu, Hengshi (2022)</p>\n<p>\"Recent developments in deep learning have enabled generation of novel and realistic images or sentences from low-dimensional representations. In addition, a revolution in biotechnology has enabled high-throughput measurement of gene expression in thousands to millions of single cells.\"</p>\n<p>\"Several deep generative models have been developed to learn latent representations of cells and generate realistic high-dimensional single-cell data. However, these deep generative models primarily generate data similar to that seen during training and have limited ability to predict gene expression of unseen cell states. In consequence, it constrains the applicability of deep generative models for single-cell data, which usually have a relatively small set of observed conditions. \"</p>\n<p>In Chapter II, the author studied two main classes of deep generative models for single-cell RNA-seq data: variational autoencoders (VAEs) and generative adversarial networks (GANs). He systematically assessed their disentanglement and generation performance and show that VAEs excel at learning cellular representations, while GANs excel at generating realistic single-cell gene expression data.\"</p>\n<p>\"He also developed MichiGAN, a novel neural network architecture that combines the strengths of VAEs and GANs to sample from disentangled representations without sacrificing data generation quality.\"</p>\n<p>\"In Chapter III, he developed PerturbNet, a novel deep generative model to generate single-cell data under unseen drug treatments. Existing approaches attempt to learn drug effects independently of cell state and cannot predict results for unseen drug treatments. To address these limitations, our PerturbNet framework learns mapping from a continuous representation of drug treatment to cellular states. PerturbNet can then generate single-cell data for both observed and unseen drug treatments. He showed that PerturbNet accurately predicts single-cell RNA-seq data resulting from unseen drug treatments. He also fine-tune PerturbNet using cellular properties to improve the continuous representations of drug treatments.\"</p>\n<p>\"In Chapter IV, he extend PerturbNet to learn single-cell responses to genetic perturbations, including pooled CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) genetic inactivations and genetic mutations. Existing approaches attempt to learn genetic perturbation effects independently of cell state and rely on one-hot encodings of genetic perturbations. He developed a GenotypeVAE model and also employ a state-of-the-art protein sequence embedding model to encode genetic perturbations into continuous representations, allowing prediction for both unseen genes and unseen gene combinations.\"</p>\n<p>\"In Chapter V, he extended PerturbNet to design optimal perturbations and attribute perturbation outcomes to specific perturbation features. He considered the translation of a group of cells to a target cell state, and propose two algorithms to design perturbations that achieve this desired target cell state. He showed that the algorithms are effective at designing perturbations that achieve the cell state translation of interest. He also employed model interpretability methods to attribute the effects of chemical or genetic perturbations to specific atoms or gene functional annotations.\"</p>\n<p><a href=\"https://deepblue.lib.umich.edu/handle/2027.42/174555\" target=\"_blank\">https://deepblue.lib.umich.edu/handle/2027.42/174555</a></p>\n<h1>Harmonized Single-Cell Perturbation Data</h1>\n<p>scPerturb: Information Resource for Harmonized Single-Cell Perturbation Data</p>\n<p>Authors: Stefan Peidli, Tessa D. Green, Ciyue Shen, Torsten Gross, Joseph Min, Jake P. Taylor-King, Debora S. Marks, Augustin Luna, Nils Blüthgen, Chris Sander<br>\ndoi: <a href=\"https://doi.org/10.1101/2022.08.20.504663\" target=\"_blank\">https://doi.org/10.1101/2022.08.20.504663</a></p>\n<p>\"Recent biotechnological advances led to growing numbers of single-cell studies, which reveal molecular and phenotypic responses to large numbers of perturbations. However, analysis across diverse datasets is typically hampered by differences in format, naming conventions, data filtering and normalization. In order to facilitate development and benchmarking of computational methods in systems biology, the authors collected a set of 44 publicly available single-cell perturbation-response datasets with molecular readouts, including transcriptomics, proteomics and epigenomics.\"</p>\n<p>\"They applied uniform pre-processing and quality control pipelines and harmonize feature annotations. The resulting information resource enables efficient development and testing of computational analysis methods, and facilitates direct comparison and integration across datasets. Using these datasets, they demonstrated the application of E-distance for quantifying perturbation similarity and strength.\" </p>\n<p><a href=\"https://www.biorxiv.org/content/10.1101/2022.08.20.504663v1.article-metrics\" target=\"_blank\">https://www.biorxiv.org/content/10.1101/2022.08.20.504663v1.article-metrics</a></p>\n<h1>scPerturb: A resource and a python tool for single-cell perturbation data</h1>\n<p>pip install scperturb</p>\n<p>E-distances<br>\nestats = edist(adata, obs_key='perturbation')<br>\n E-distances to a specific group (e.g. 'control')<br>\nestats_control = estats.loc['control']<br>\n E-test for difference to control<br>\ndf = etest(adata, obs_key='perturbation', obsm_key='X_pca', dist='sqeuclidean', control='control', alpha=0.05, runs=100)</p>\n<p><a href=\"https://github.com/sanderlab/scPerturb\" target=\"_blank\">https://github.com/sanderlab/scPerturb</a><br>\n<a href=\"http://projects.sanderlab.org/scperturb/\" target=\"_blank\">http://projects.sanderlab.org/scperturb/</a></p>",
      "rawMarkdown": "#PerturbNet, MichiGAN\n\nDeep Generative Models for Single-Cell Perturbation Experiments\n\nAuthor: Yu, Hengshi (2022)\n\n\"Recent developments in deep learning have enabled generation of novel and realistic images or sentences from low-dimensional representations. In addition, a revolution in biotechnology has enabled high-throughput measurement of gene expression in thousands to millions of single cells.\"\n\n\"Several deep generative models have been developed to learn latent representations of cells and generate realistic high-dimensional single-cell data. However, these deep generative models primarily generate data similar to that seen during training and have limited ability to predict gene expression of unseen cell states. In consequence, it constrains the applicability of deep generative models for single-cell data, which usually have a relatively small set of observed conditions. \"\n\nIn Chapter II, the author studied two main classes of deep generative models for single-cell RNA-seq data: variational autoencoders (VAEs) and generative adversarial networks (GANs). He systematically assessed their disentanglement and generation performance and show that VAEs excel at learning cellular representations, while GANs excel at generating realistic single-cell gene expression data.\"\n\n\"He also developed MichiGAN, a novel neural network architecture that combines the strengths of VAEs and GANs to sample from disentangled representations without sacrificing data generation quality.\"\n\n\"In Chapter III, he developed PerturbNet, a novel deep generative model to generate single-cell data under unseen drug treatments. Existing approaches attempt to learn drug effects independently of cell state and cannot predict results for unseen drug treatments. To address these limitations, our PerturbNet framework learns mapping from a continuous representation of drug treatment to cellular states. PerturbNet can then generate single-cell data for both observed and unseen drug treatments. He showed that PerturbNet accurately predicts single-cell RNA-seq data resulting from unseen drug treatments. He also fine-tune PerturbNet using cellular properties to improve the continuous representations of drug treatments.\"\n\n\"In Chapter IV, he extend PerturbNet to learn single-cell responses to genetic perturbations, including pooled CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) genetic inactivations and genetic mutations. Existing approaches attempt to learn genetic perturbation effects independently of cell state and rely on one-hot encodings of genetic perturbations. He developed a GenotypeVAE model and also employ a state-of-the-art protein sequence embedding model to encode genetic perturbations into continuous representations, allowing prediction for both unseen genes and unseen gene combinations.\"\n\n\"In Chapter V, he extended PerturbNet to design optimal perturbations and attribute perturbation outcomes to specific perturbation features. He considered the translation of a group of cells to a target cell state, and propose two algorithms to design perturbations that achieve this desired target cell state. He showed that the algorithms are effective at designing perturbations that achieve the cell state translation of interest. He also employed model interpretability methods to attribute the effects of chemical or genetic perturbations to specific atoms or gene functional annotations.\"\n\nhttps://deepblue.lib.umich.edu/handle/2027.42/174555\n\n# Harmonized Single-Cell Perturbation Data\n\nscPerturb: Information Resource for Harmonized Single-Cell Perturbation Data\n\nAuthors: Stefan Peidli, Tessa D. Green, Ciyue Shen, Torsten Gross, Joseph Min, Jake P. Taylor-King, Debora S. Marks, Augustin Luna, Nils Blüthgen, Chris Sander\ndoi: https://doi.org/10.1101/2022.08.20.504663\n\n\"Recent biotechnological advances led to growing numbers of single-cell studies, which reveal molecular and phenotypic responses to large numbers of perturbations. However, analysis across diverse datasets is typically hampered by differences in format, naming conventions, data filtering and normalization. In order to facilitate development and benchmarking of computational methods in systems biology, the authors collected a set of 44 publicly available single-cell perturbation-response datasets with molecular readouts, including transcriptomics, proteomics and epigenomics.\"\n\n\"They applied uniform pre-processing and quality control pipelines and harmonize feature annotations. The resulting information resource enables efficient development and testing of computational analysis methods, and facilitates direct comparison and integration across datasets. Using these datasets, they demonstrated the application of E-distance for quantifying perturbation similarity and strength.\" \n\nhttps://www.biorxiv.org/content/10.1101/2022.08.20.504663v1.article-metrics\n\n#scPerturb: A resource and a python tool for single-cell perturbation data\n\npip install scperturb\n\n E-distances\nestats = edist(adata, obs_key='perturbation')\n E-distances to a specific group (e.g. 'control')\nestats_control = estats.loc['control']\n E-test for difference to control\ndf = etest(adata, obs_key='perturbation', obsm_key='X_pca', dist='sqeuclidean', control='control', alpha=0.05, runs=100)\n\nhttps://github.com/sanderlab/scPerturb\nhttp://projects.sanderlab.org/scperturb/",
      "votes": null
    },
    {
      "id": "2435305",
      "postDate": "09/12/2023 21:01:53",
      "content": "<p>GEARS from  Stanford - take a look <br>\n<a href=\"https://youtu.be/UotniZ-gxU0?t=4351\" target=\"_blank\">https://youtu.be/UotniZ-gxU0?t=4351</a><br>\n<a href=\"https://www.nature.com/articles/s41467-022-32111-8?s=35\" target=\"_blank\">https://www.nature.com/articles/s41467-022-32111-8?s=35</a><br>\n<a href=\"https://twitter.com/jure/status/1549086373587521536?t=rRxQkjfPuMzV5LTSsuEvxg&amp;s=09\" target=\"_blank\">https://twitter.com/jure/status/1549086373587521536?t=rRxQkjfPuMzV5LTSsuEvxg&amp;s=09</a></p>\n<p>Also <br>\nKaggle competition \"MoA\" </p>",
      "rawMarkdown": "GEARS from  Stanford - take a look \nhttps://youtu.be/UotniZ-gxU0?t=4351\nhttps://www.nature.com/articles/s41467-022-32111-8?s=35\nhttps://twitter.com/jure/status/1549086373587521536?t=rRxQkjfPuMzV5LTSsuEvxg&s=09\n\nAlso \nKaggle competition \"MoA\"",
      "votes": null
    },
    {
      "id": "2435444",
      "postDate": "09/13/2023 01:52:37",
      "content": "<p>Useful introductory notebook from <a href=\"https://www.sc-best-practices.org/conditions/perturbation_modeling.html\" target=\"_blank\">sc-best-practices</a></p>",
      "rawMarkdown": "Useful introductory notebook from [sc-best-practices](https://www.sc-best-practices.org/conditions/perturbation_modeling.html)",
      "votes": null
    },
    {
      "id": "2435668",
      "postDate": "09/13/2023 06:01:06",
      "content": "<p>it is a useful and informative post. thanks for sharing👏</p>",
      "rawMarkdown": "it is a useful and informative post. thanks for sharing👏",
      "votes": null
    },
    {
      "id": "2436261",
      "postDate": "09/13/2023 13:10:54",
      "content": "<p>Yes I read those control perturbations on MoA competition. Though they are more related to drugs which makes it harder for me to understand.  There are very few material about Perturbation on Kaggle till now.</p>\n<p>Thank you for the tips Chervov. Your expertise in Bioinformatics is extremely valuable for all of us.</p>",
      "rawMarkdown": "Yes I read those control perturbations on MoA competition. Though they are more related to drugs which makes it harder for me to understand.  There are very few material about Perturbation on Kaggle till now.\n\nThank you for the tips Chervov. Your expertise in Bioinformatics is extremely valuable for all of us.",
      "votes": null
    },
    {
      "id": "2436262",
      "postDate": "09/13/2023 13:12:13",
      "content": "<p>Thank you Kane for the nice words and support.</p>",
      "rawMarkdown": "Thank you Kane for the nice words and support.",
      "votes": null
    },
    {
      "id": "2436265",
      "postDate": "09/13/2023 13:13:55",
      "content": "<p>Thank you for your appreciation Mehmet. It means a lot for me.</p>",
      "rawMarkdown": "Thank you for your appreciation Mehmet. It means a lot for me.",
      "votes": null
    },
    {
      "id": "2436270",
      "postDate": "09/13/2023 13:17:34",
      "content": "<p>your work is worth it, admirable👏</p>",
      "rawMarkdown": "your work is worth it, admirable👏",
      "votes": null
    },
    {
      "id": "2439767",
      "postDate": "09/15/2023 05:41:00",
      "content": "<p>Informative content <a href=\"https://www.kaggle.com/mpwolke\" target=\"_blank\">@mpwolke</a>. Thanks for sharing!</p>",
      "rawMarkdown": "Informative content @mpwolke. Thanks for sharing!",
      "votes": null
    },
    {
      "id": "2440436",
      "postDate": "09/15/2023 14:05:58",
      "content": "<p>Thank you for your nice words Sudheesh.</p>",
      "rawMarkdown": "Thank you for your nice words Sudheesh.",
      "votes": null
    },
    {
      "id": "2444600",
      "postDate": "09/18/2023 11:07:17",
      "content": "<p>Thanks for the information. It was very much informatibe</p>",
      "rawMarkdown": "Thanks for the information. It was very much informatibe",
      "votes": null
    },
    {
      "id": "2477146",
      "postDate": "10/11/2023 06:01:11",
      "content": "<p>Thank you <a href=\"https://www.kaggle.com/alexandervc\" target=\"_blank\">@alexandervc</a> </p>",
      "rawMarkdown": "Thank you @alexandervc",
      "votes": null
    },
    {
      "id": "2477147",
      "postDate": "10/11/2023 06:02:13",
      "content": "<p>Thanks <a href=\"https://www.kaggle.com/mpwolke\" target=\"_blank\">@mpwolke</a> for sharing useful information.</p>",
      "rawMarkdown": "Thanks @mpwolke for sharing useful information.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2435305,
      "author_name": "alexandervc",
      "author_url": "",
      "post_date": "09/12/2023 21:01:53",
      "content": "<p>GEARS from  Stanford - take a look <br>\n<a href=\"https://youtu.be/UotniZ-gxU0?t=4351\" target=\"_blank\">https://youtu.be/UotniZ-gxU0?t=4351</a><br>\n<a href=\"https://www.nature.com/articles/s41467-022-32111-8?s=35\" target=\"_blank\">https://www.nature.com/articles/s41467-022-32111-8?s=35</a><br>\n<a href=\"https://twitter.com/jure/status/1549086373587521536?t=rRxQkjfPuMzV5LTSsuEvxg&amp;s=09\" target=\"_blank\">https://twitter.com/jure/status/1549086373587521536?t=rRxQkjfPuMzV5LTSsuEvxg&amp;s=09</a></p>\n<p>Also <br>\nKaggle competition \"MoA\" </p>",
      "votes": null,
      "replies": [
        {
          "id": 2436261,
          "author_name": "mpwolke",
          "author_url": "",
          "post_date": "09/13/2023 13:10:54",
          "content": "<p>Yes I read those control perturbations on MoA competition. Though they are more related to drugs which makes it harder for me to understand.  There are very few material about Perturbation on Kaggle till now.</p>\n<p>Thank you for the tips Chervov. Your expertise in Bioinformatics is extremely valuable for all of us.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 2477146,
          "author_name": "pranavbelhekar",
          "author_url": "",
          "post_date": "10/11/2023 06:01:11",
          "content": "<p>Thank you <a href=\"https://www.kaggle.com/alexandervc\" target=\"_blank\">@alexandervc</a> </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2435444,
      "author_name": "kane9530",
      "author_url": "",
      "post_date": "09/13/2023 01:52:37",
      "content": "<p>Useful introductory notebook from <a href=\"https://www.sc-best-practices.org/conditions/perturbation_modeling.html\" target=\"_blank\">sc-best-practices</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 2436262,
          "author_name": "mpwolke",
          "author_url": "",
          "post_date": "09/13/2023 13:12:13",
          "content": "<p>Thank you Kane for the nice words and support.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2435668,
      "author_name": "mehmetisik",
      "author_url": "",
      "post_date": "09/13/2023 06:01:06",
      "content": "<p>it is a useful and informative post. thanks for sharing👏</p>",
      "votes": null,
      "replies": [
        {
          "id": 2436265,
          "author_name": "mpwolke",
          "author_url": "",
          "post_date": "09/13/2023 13:13:55",
          "content": "<p>Thank you for your appreciation Mehmet. It means a lot for me.</p>",
          "votes": null,
          "replies": [
            {
              "id": 2436270,
              "author_name": "mehmetisik",
              "author_url": "",
              "post_date": "09/13/2023 13:17:34",
              "content": "<p>your work is worth it, admirable👏</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2439767,
      "author_name": "sudheeshr",
      "author_url": "",
      "post_date": "09/15/2023 05:41:00",
      "content": "<p>Informative content <a href=\"https://www.kaggle.com/mpwolke\" target=\"_blank\">@mpwolke</a>. Thanks for sharing!</p>",
      "votes": null,
      "replies": [
        {
          "id": 2440436,
          "author_name": "mpwolke",
          "author_url": "",
          "post_date": "09/15/2023 14:05:58",
          "content": "<p>Thank you for your nice words Sudheesh.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2444600,
      "author_name": "laxminarayanasahu",
      "author_url": "",
      "post_date": "09/18/2023 11:07:17",
      "content": "<p>Thanks for the information. It was very much informatibe</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2477147,
      "author_name": "pranavbelhekar",
      "author_url": "",
      "post_date": "10/11/2023 06:02:13",
      "content": "<p>Thanks <a href=\"https://www.kaggle.com/mpwolke\" target=\"_blank\">@mpwolke</a> for sharing useful information.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2435272": "#PerturbNet, MichiGAN\n\nDeep Generative Models for Single-Cell Perturbation Experiments\n\nAuthor: Yu, Hengshi (2022)\n\n\"Recent developments in deep learning have enabled generation of novel and realistic images or sentences from low-dimensional representations. In addition, a revolution in biotechnology has enabled high-throughput measurement of gene expression in thousands to millions of single cells.\"\n\n\"Several deep generative models have been developed to learn latent representations of cells and generate realistic high-dimensional single-cell data. However, these deep generative models primarily generate data similar to that seen during training and have limited ability to predict gene expression of unseen cell states. In consequence, it constrains the applicability of deep generative models for single-cell data, which usually have a relatively small set of observed conditions. \"\n\nIn Chapter II, the author studied two main classes of deep generative models for single-cell RNA-seq data: variational autoencoders (VAEs) and generative adversarial networks (GANs). He systematically assessed their disentanglement and generation performance and show that VAEs excel at learning cellular representations, while GANs excel at generating realistic single-cell gene expression data.\"\n\n\"He also developed MichiGAN, a novel neural network architecture that combines the strengths of VAEs and GANs to sample from disentangled representations without sacrificing data generation quality.\"\n\n\"In Chapter III, he developed PerturbNet, a novel deep generative model to generate single-cell data under unseen drug treatments. Existing approaches attempt to learn drug effects independently of cell state and cannot predict results for unseen drug treatments. To address these limitations, our PerturbNet framework learns mapping from a continuous representation of drug treatment to cellular states. PerturbNet can then generate single-cell data for both observed and unseen drug treatments. He showed that PerturbNet accurately predicts single-cell RNA-seq data resulting from unseen drug treatments. He also fine-tune PerturbNet using cellular properties to improve the continuous representations of drug treatments.\"\n\n\"In Chapter IV, he extend PerturbNet to learn single-cell responses to genetic perturbations, including pooled CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) genetic inactivations and genetic mutations. Existing approaches attempt to learn genetic perturbation effects independently of cell state and rely on one-hot encodings of genetic perturbations. He developed a GenotypeVAE model and also employ a state-of-the-art protein sequence embedding model to encode genetic perturbations into continuous representations, allowing prediction for both unseen genes and unseen gene combinations.\"\n\n\"In Chapter V, he extended PerturbNet to design optimal perturbations and attribute perturbation outcomes to specific perturbation features. He considered the translation of a group of cells to a target cell state, and propose two algorithms to design perturbations that achieve this desired target cell state. He showed that the algorithms are effective at designing perturbations that achieve the cell state translation of interest. He also employed model interpretability methods to attribute the effects of chemical or genetic perturbations to specific atoms or gene functional annotations.\"\n\nhttps://deepblue.lib.umich.edu/handle/2027.42/174555\n\n# Harmonized Single-Cell Perturbation Data\n\nscPerturb: Information Resource for Harmonized Single-Cell Perturbation Data\n\nAuthors: Stefan Peidli, Tessa D. Green, Ciyue Shen, Torsten Gross, Joseph Min, Jake P. Taylor-King, Debora S. Marks, Augustin Luna, Nils Blüthgen, Chris Sander\ndoi: https://doi.org/10.1101/2022.08.20.504663\n\n\"Recent biotechnological advances led to growing numbers of single-cell studies, which reveal molecular and phenotypic responses to large numbers of perturbations. However, analysis across diverse datasets is typically hampered by differences in format, naming conventions, data filtering and normalization. In order to facilitate development and benchmarking of computational methods in systems biology, the authors collected a set of 44 publicly available single-cell perturbation-response datasets with molecular readouts, including transcriptomics, proteomics and epigenomics.\"\n\n\"They applied uniform pre-processing and quality control pipelines and harmonize feature annotations. The resulting information resource enables efficient development and testing of computational analysis methods, and facilitates direct comparison and integration across datasets. Using these datasets, they demonstrated the application of E-distance for quantifying perturbation similarity and strength.\" \n\nhttps://www.biorxiv.org/content/10.1101/2022.08.20.504663v1.article-metrics\n\n#scPerturb: A resource and a python tool for single-cell perturbation data\n\npip install scperturb\n\n E-distances\nestats = edist(adata, obs_key='perturbation')\n E-distances to a specific group (e.g. 'control')\nestats_control = estats.loc['control']\n E-test for difference to control\ndf = etest(adata, obs_key='perturbation', obsm_key='X_pca', dist='sqeuclidean', control='control', alpha=0.05, runs=100)\n\nhttps://github.com/sanderlab/scPerturb\nhttp://projects.sanderlab.org/scperturb/",
    "2435305": "GEARS from  Stanford - take a look \nhttps://youtu.be/UotniZ-gxU0?t=4351\nhttps://www.nature.com/articles/s41467-022-32111-8?s=35\nhttps://twitter.com/jure/status/1549086373587521536?t=rRxQkjfPuMzV5LTSsuEvxg&s=09\n\nAlso \nKaggle competition \"MoA\"",
    "2435444": "Useful introductory notebook from [sc-best-practices](https://www.sc-best-practices.org/conditions/perturbation_modeling.html)",
    "2435668": "it is a useful and informative post. thanks for sharing👏",
    "2436261": "Yes I read those control perturbations on MoA competition. Though they are more related to drugs which makes it harder for me to understand.  There are very few material about Perturbation on Kaggle till now.\n\nThank you for the tips Chervov. Your expertise in Bioinformatics is extremely valuable for all of us.",
    "2436262": "Thank you Kane for the nice words and support.",
    "2436265": "Thank you for your appreciation Mehmet. It means a lot for me.",
    "2436270": "your work is worth it, admirable👏",
    "2439767": "Informative content @mpwolke. Thanks for sharing!",
    "2440436": "Thank you for your nice words Sudheesh.",
    "2444600": "Thanks for the information. It was very much informatibe",
    "2477146": "Thank you @alexandervc",
    "2477147": "Thanks @mpwolke for sharing useful information."
  },
  "source": "meta"
}