Linearly decoded VAE (LDVAE) is a generative model similar to PCA which can be used similarly as scVI, however, due that it has linear functions instead of neural networks it's more easy to interpret its ouput. In this notebook i used this model to get differential expression analysis.
\nSCMER (single-cell manifold-preserving feature
\nselection) was used in notebook to get a compact version of the full dataset. I also do some visualizations and i have to admit that is too easy to use with Scanpy, once you have your data in form of adata object.
After some practice with SCMER i decided to use it to get a matrix that i could handle to train graph neural networks and only 13GB RAM. The GNN used is scGAE, a recently published model on Nature. scGAE performs clustering on its latent space and it can be used for downstram analysis.
\nIn this notebook, i used TabNet, a kind of transformer useful for tabular data. As many now, chromatin accessibility data is just a matrix. This matrix is contained on my adata object, adata.X. In resume, i used the mentioned matrix and adata.obs[\"nucleosome_signal\"] as my labels, i split it, trained my model and calculate feature importance.
Here i used for the first time EpiScanpy and PeakVi. The can be used so easy on this kind of data.
", "messages": [ { "id": 1901567, "postDate": "2022-08-16T18:45:47.290Z", "content": "Linearly decoded VAE (LDVAE) is a generative model similar to PCA which can be used similarly as scVI, however, due that it has linear functions instead of neural networks it's more easy to interpret its ouput. In this notebook i used this model to get differential expression analysis.
\nSCMER (single-cell manifold-preserving feature
\nselection) was used in notebook to get a compact version of the full dataset. I also do some visualizations and i have to admit that is too easy to use with Scanpy, once you have your data in form of adata object.
After some practice with SCMER i decided to use it to get a matrix that i could handle to train graph neural networks and only 13GB RAM. The GNN used is scGAE, a recently published model on Nature. scGAE performs clustering on its latent space and it can be used for downstram analysis.
\nIn this notebook, i used TabNet, a kind of transformer useful for tabular data. As many now, chromatin accessibility data is just a matrix. This matrix is contained on my adata object, adata.X. In resume, i used the mentioned matrix and adata.obs[\"nucleosome_signal\"] as my labels, i split it, trained my model and calculate feature importance.
Here i used for the first time EpiScanpy and PeakVi. The can be used so easy on this kind of data.
", "rawMarkdown": "# scRNA-seq 🧬\n\n## 1. [scRNA-seq 🧬: Differential Expression with scVI](https://www.kaggle.com/hiramcho/scrna-seq-differential-expression-with-scvi) \n**Linearly decoded VAE** ([LDVAE](https://docs.scvi-tools.org/en/stable/tutorials/notebooks/linear_decoder.html)) is a generative model similar to PCA which can be used similarly as [scVI](https://docs.scvi-tools.org/en/stable/tutorials/notebooks/api_overview.html), however, due that it has linear functions instead of neural networks it's more easy to interpret its ouput. In this notebook i used this model to get **differential expression analysis**. \n\n\n## 2. [scRNA-seq 🧬: Scanpy & SCMER for Feature Selection](https://www.kaggle.com/hiramcho/scrna-seq-scanpy-scmer-for-feature-selection)\n\n[SCMER](https://scmer.readthedocs.io/en/latest/) (single-cell manifold-preserving feature\nselection) was used in notebook to get a compact version of the full dataset. I also do some visualizations and i have to admit that is too easy to use with [Scanpy](https://scanpy.readthedocs.io/en/stable/index.html), once you have your data in form of `adata` object.\n\n\n## 3. [scRNA-seq 🧬: scGAE with Spektral and RAPIDS](https://www.kaggle.com/hiramcho/scrna-seq-scgae-with-spektral-and-rapids)\n\nAfter some practice with SCMER i decided to use it to get a matrix that i could handle to train **graph neural networks** and only 13GB RAM. The GNN used is [scGAE](https://www.nature.com/articles/s41598-021-99003-7), a recently published model on Nature. scGAE performs clustering on its **latent space** and it can be used for downstram analysis. \n\n\n# scATAC-seq 🧬\n\n## 4. [scATAC-seq 🧬: Feature Importance with TabNet](https://www.kaggle.com/hiramcho/scatac-seq-feature-importance-with-tabnet) \nIn this notebook, i used **TabNet**, a kind of transformer useful for tabular data. As many now, **chromatin accessibility data** is just a *matrix*. This matrix is contained on my `adata` object, `adata.X`. In resume, i used the mentioned matrix and `adata.obs[\"nucleosome_signal\"]` as my labels, i split it, trained my model and calculate **feature importance**. \n\n\n## 5. [scATAC-seq 🧬: EpiScanpy & PeakVI](https://www.kaggle.com/code/hiramcho/scatac-seq-episcanpy-peakvi) \n\nHere i used for the first time [EpiScanpy](https://episcanpy.readthedocs.io/) and [PeakVi](https://docs.scvi-tools.org/en/stable/tutorials/notebooks/PeakVI.html). The can be used so easy on this kind of data. ", "votes": 37 }, { "id": 1917923, "postDate": "2022-08-29T06:52:53.210Z", "content": "Hi @hiramcho thanks for sharing the ideas！
", "rawMarkdown": "Hi @hiramcho thanks for sharing the ideas！", "votes": 1 }, { "id": 1916332, "postDate": "2022-08-27T19:14:07.823Z", "content": "@hiramcho Please look at:
\nhttps://www.kaggle.com/competitions/open-problems-multimodal/discussion/348293
\nhttps://www.kaggle.com/competitions/open-problems-multimodal/discussion/348294
\nWould be so kind to join and/or give a webinar on your work ?
Hi, I've never joined to a webinar, if you help me sharing what do you need from me it would be awesome.
", "rawMarkdown": "Hi, I've never joined to a webinar, if you help me sharing what do you need from me it would be awesome." }, { "id": 1916424, "postDate": "2022-08-27T21:29:16.523Z", "content": "Just explain about your notebooks or something related + explain some biological background,
\nwhat biological question is behind , what was known before, etc..
\nFor example If it can be applied to current competition (up to you).
\nJust standard presentation of the work…
Sure, when is this webinar?
", "rawMarkdown": "Sure, when is this webinar?", "votes": 1 }, { "id": 1916825, "postDate": "2022-08-28T07:29:20.483Z", "content": "Great ! Thank you !
\nThere is already planned webinars on Monday and Wednesday (see link post above).
\nMay be you can give webinar - the next Monday - 5 September ?
\nOr Tuesday - 6 or Wednesday - 7? What is convenient for you ?
Sure, It would a pleasure
", "rawMarkdown": "Sure, It would a pleasure\n", "votes": 2 }, { "id": 1917536, "postDate": "2022-08-28T20:18:01.460Z", "content": "So Monday 5 September is Okay, yes ?
\nWhat is your time zone ?
\nTime zone of Paris - evening time say 18.00 is it Okay ? Or 17.00 or 19.00 (that is typical webinar time in our community) .
It's ok, I think that that time is appropiate
", "rawMarkdown": "It's ok, I think that that time is appropiate", "votes": 1 }, { "id": 1918709, "postDate": "2022-08-29T19:06:46.890Z", "content": "Great ! So planned !
\nLet us keep in contact !
@hiramcho
\nHello Hiram,
\njust to confirm - is everything Okay with your webinar on Monday ? (which we discussed before)
Thanks for sharing!
", "rawMarkdown": "Thanks for sharing! " }, { "id": 2028799, "postDate": "2022-11-14T08:22:58.823Z", "content": "Thank you very much for sharing this!
", "rawMarkdown": "Thank you very much for sharing this!\t\t" }, { "id": 2028776, "postDate": "2022-11-14T08:11:17.377Z", "content": "This is amazing!Thanks
", "rawMarkdown": "This is amazing!Thanks" } ], "comments": [ { "id": 1917923, "author_name": "A Beginner", "author_url": "", "post_date": "2022-08-29T06:52:53.210000", "content": "Hi @hiramcho thanks for sharing the ideas！
", "votes": 1, "replies": [] }, { "id": 1916332, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2022-08-27T19:14:07.823000", "content": "@hiramcho Please look at:
\nhttps://www.kaggle.com/competitions/open-problems-multimodal/discussion/348293
\nhttps://www.kaggle.com/competitions/open-problems-multimodal/discussion/348294
\nWould be so kind to join and/or give a webinar on your work ?
Hi, I've never joined to a webinar, if you help me sharing what do you need from me it would be awesome.
", "votes": 0, "replies": [] }, { "id": 1916424, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2022-08-27T21:29:16.523000", "content": "Just explain about your notebooks or something related + explain some biological background,
\nwhat biological question is behind , what was known before, etc..
\nFor example If it can be applied to current competition (up to you).
\nJust standard presentation of the work…
Sure, when is this webinar?
", "votes": 1, "replies": [] }, { "id": 1916825, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2022-08-28T07:29:20.483000", "content": "Great ! Thank you !
\nThere is already planned webinars on Monday and Wednesday (see link post above).
\nMay be you can give webinar - the next Monday - 5 September ?
\nOr Tuesday - 6 or Wednesday - 7? What is convenient for you ?
Sure, It would a pleasure
", "votes": 2, "replies": [] }, { "id": 1917536, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2022-08-28T20:18:01.460000", "content": "So Monday 5 September is Okay, yes ?
\nWhat is your time zone ?
\nTime zone of Paris - evening time say 18.00 is it Okay ? Or 17.00 or 19.00 (that is typical webinar time in our community) .
It's ok, I think that that time is appropiate
", "votes": 1, "replies": [] }, { "id": 1918709, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2022-08-29T19:06:46.890000", "content": "Great ! So planned !
\nLet us keep in contact !
@hiramcho
\nHello Hiram,
\njust to confirm - is everything Okay with your webinar on Monday ? (which we discussed before)
Thanks for sharing!
", "votes": 0, "replies": [] }, { "id": 2028799, "author_name": "", "author_url": "", "post_date": "2022-11-14T08:22:58.823000", "content": "Thank you very much for sharing this!
", "votes": 0, "replies": [] }, { "id": 2028776, "author_name": "", "author_url": "", "post_date": "2022-11-14T08:11:17.377000", "content": "This is amazing!Thanks
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1901567": "# scRNA-seq 🧬\n\n## 1. [scRNA-seq 🧬: Differential Expression with scVI](https://www.kaggle.com/hiramcho/scrna-seq-differential-expression-with-scvi) \n**Linearly decoded VAE** ([LDVAE](https://docs.scvi-tools.org/en/stable/tutorials/notebooks/linear_decoder.html)) is a generative model similar to PCA which can be used similarly as [scVI](https://docs.scvi-tools.org/en/stable/tutorials/notebooks/api_overview.html), however, due that it has linear functions instead of neural networks it's more easy to interpret its ouput. In this notebook i used this model to get **differential expression analysis**. \n\n\n## 2. [scRNA-seq 🧬: Scanpy & SCMER for Feature Selection](https://www.kaggle.com/hiramcho/scrna-seq-scanpy-scmer-for-feature-selection)\n\n[SCMER](https://scmer.readthedocs.io/en/latest/) (single-cell manifold-preserving feature\nselection) was used in notebook to get a compact version of the full dataset. I also do some visualizations and i have to admit that is too easy to use with [Scanpy](https://scanpy.readthedocs.io/en/stable/index.html), once you have your data in form of `adata` object.\n\n\n## 3. [scRNA-seq 🧬: scGAE with Spektral and RAPIDS](https://www.kaggle.com/hiramcho/scrna-seq-scgae-with-spektral-and-rapids)\n\nAfter some practice with SCMER i decided to use it to get a matrix that i could handle to train **graph neural networks** and only 13GB RAM. The GNN used is [scGAE](https://www.nature.com/articles/s41598-021-99003-7), a recently published model on Nature. scGAE performs clustering on its **latent space** and it can be used for downstram analysis. \n\n\n# scATAC-seq 🧬\n\n## 4. [scATAC-seq 🧬: Feature Importance with TabNet](https://www.kaggle.com/hiramcho/scatac-seq-feature-importance-with-tabnet) \nIn this notebook, i used **TabNet**, a kind of transformer useful for tabular data. As many now, **chromatin accessibility data** is just a *matrix*. This matrix is contained on my `adata` object, `adata.X`. In resume, i used the mentioned matrix and `adata.obs[\"nucleosome_signal\"]` as my labels, i split it, trained my model and calculate **feature importance**. \n\n\n## 5. [scATAC-seq 🧬: EpiScanpy & PeakVI](https://www.kaggle.com/code/hiramcho/scatac-seq-episcanpy-peakvi) \n\nHere i used for the first time [EpiScanpy](https://episcanpy.readthedocs.io/) and [PeakVi](https://docs.scvi-tools.org/en/stable/tutorials/notebooks/PeakVI.html). The can be used so easy on this kind of data. ", "1917923": "Hi @hiramcho thanks for sharing the ideas！", "1916332": "@hiramcho Please look at:\nhttps://www.kaggle.com/competitions/open-problems-multimodal/discussion/348293\nhttps://www.kaggle.com/competitions/open-problems-multimodal/discussion/348294\nWould be so kind to join and/or give a webinar on your work ? ", "1924608": "@hiramcho \nHello Hiram,\njust to confirm - is everything Okay with your webinar on Monday ? (which we discussed before) ", "2014777": "", "2167376": "Thanks for sharing! ", "2028799": "Thank you very much for sharing this!\t\t", "2028776": "This is amazing!Thanks" } }