Lisbon single cell workshop 2021

This is a repository for the notebooks used in the Single Cell workshop for the Lisbon Summer School 2021. You can also see/use these notebooks in Google Colab.

To be able to run this in Colab, you will need to do the following:

• Open a Google account: If you have a personal Gmail account you are good to go. Otherwise go here and register for one. If you don't want to use your pesonal account, you can register using your university account as well.

• Set up GDrive: Each Google Account comes with 15 GiB of storage free to use. You can save your data there.

• Get Data: You can download the data from the Heart Cell Atlas website. For the batch correction excercise you will need the Heart Immune Cells, and for the label transfer you will need the heart fibroblast.

For the transfer label example using scNym we can either train the model on the whole heart cell atlas dataset, but this may take a while even with GPUs! To speed up things, you can save the heart model (models/heart_scnym) in your GDrive so it can be used to annotate the dummy dataset.

Correcting batches for single cell data (scRNA-Seq)

To this end, we will use scvi-tools. You can install the suite in your local machine (Linux/MacOS) with pip install scvi-tools. However, as with many deep learning workflows, access to a GPU will dramatically speed up things.

You can access the notebook here in Colab to run the example to batch-correct cardiac immune cells. Make a copy of it and save it in your own GDrive.

To use it you will need to follow these steps:

• First to download the scRNA-Seq raw counts for the cardiac immune cells directly from the Heart Cell Atlas portal.

• In your GDrive home directory create the folders needed for the following path: /INBOX/lisbon_2021/raw_data/. You can also create your own path.

• Upload your anndata object inside /INBOX/lisbon_2021/raw_data/.

• Update the path to the object in the CoLab notebook. It is better if you make a copy of the notebook into your own GDrive!!!

You are good to go!

Transferring labels between single cell data (scRNA-Seq) using scNym

Here we will use scNym which uses semi-supervised adversarial neural networks to tarnsfer labels between a reference and a query single cell data. It is intended to be used with scRNA-Seq, but you can be adventurous and tried with other dfata modalities.

You can access the notebook here in Colab.

For the purpose of this course, the trainig step has already been done for you. You can access the model files in this repo under the models folder. For demonstration purposes, we will be using the cardiac fibroblast datasets from the heart cell atlas. But you are strongly encouraged to try with other datasets. Please follow these steps:

• First to download the scRNA-Seq raw counts for the cardiac fibroblasts directly from the Heart Cell Atlas portal.

• Save the object inside the /INBOX/lisbon_2021/raw_data/ path.

• Update the path to the object in the CoLab notebook. It is better if you make a copy of the notebook into your own GDrive!!!

You can later upload other public data or your own dataset as well.