gibsramen/CLAIM-scAEs

A tutorial for using an autoencoder neural network model to create a visually interpretable latent space of single cell expression data.

Autoencoder Workshop

Challenge

Use an autoencoder neural network model to create a visually interpretable latent space.

Learning Goals

• get hands on experience with building and training models with pytorch!
• play with autoencoders

Resources

Dataset

The dataset used here is the “PBMC3k” single-cell RNA-seq datasets; 3K Peripheral blood mononuclear cells from a healthy donor. Data was acquired from scanpy datasets. You will find the following related files in data:

• pbmc3k_raw_var_genes.tsv: processed dataset keeping filtered cells and genes
• pbmc3k_SeuratMetadata.tsv: cell type labels

Notebooks

Notebooks should be executed in the order listed.

1. Collect_Datasets_and_Preprocess.ipynb (already ran)
2. Collect_Cell_Type_Labels.ipynb (already ran)
3. Train_Autoencoder_Tutorial.ipynb (main notebook; this is what you will use to train your autoencoder)

Scripts

To keep the notebooks easy to read, functions are stored in the scripts folder. You will find:

• autoencoders.py: where we define some basic autoencoder model architectures in PyTorch
• train.py: utility functions for training the autoencoder model
• utils.py: miscellanous utility functions including visualize() for visualizing the autoencoder latent embedding layer

Getting Started

We will use NRNB compute resources for this workshop.

Environment Setup

1. Start a JupyterLab instance.

2. Activate the prepared shared conda environment on NRNB (no installs needed). Then register the environment as a new kernel in your JupyterLab install.

   source activate /path/to/env
   python -m ipykernel install --user --name ml_env --display-name "ml_env"

3. Refresh the JupyterLab interface page. You should now be able to access the ml_env kernel for the notebooks.

4. Let's get a copy of the repo:

   git clone https://github.com/adamklie/CLAIM-scAEs.git