Single-cell assisted deconvolutional network

Scaden is a deep-learning based algorithm for cell type deconvolution of bulk RNA-seq samples. It was developed at the DZNE Tübingen and the ZMNH in Hamburg. The method is published in Science Advances: Deep-learning based cell composition analysis from tissue expression profiles

A complete documentation is available here

Scaden overview. a) Generation of artificial bulk samples with known cell type composition from scRNA-seq data. b) Training of Scaden model ensemble on simulated training data. c) Scaden ensemble architecture. d) A trained Scaden model can be used to deconvolve complex bulk mixtures.

Installation guide

Scaden can be easily installed on a Linux system, and should also work on Mac. There are currently two options for installing Scaden, either using Bioconda or via pip.

pip

To install Scaden via pip, simply run the following command:

pip install scaden

Bioconda

You can also install Scaden via bioconda, using:

conda install -c bioconda scaden

GPU

If you want to make use of your GPU, you will have to additionally install tensorflow-gpu.

For pip:

pip install tensorflow-gpu

For conda:

conda install tensorflow-gpu

Docker

If you don't want to install Scaden at all, but rather use a Docker container, we provide that as well. For every release, we provide two version - one for CPU and one for GPU usage. To pull the CPU container, use this command:

docker pull ghcr.io/kevinmenden/scaden/scaden

For the GPU container:

docker pull ghcr.io/kevinmenden/scaden/scaden-gpu

Webtool (beta)

Additionally, we now proivde a web tool:

https://scaden.ims.bio

It contains pre-generated training datasets for several tissues, and all you need to do is to upload your expression data. Please note that this is still in preview.

Usage

We provide a detailed instructions for how to use Scaden at our Documentation page

A deconvolution workflow with Scaden consists of four major steps:

• data simulation
• data processing
• training
• prediction

If training data is already available, you can start at the data processing step. Otherwise you will first have to process scRNA-seq datasets and perform data simulation to generate a training dataset. As an example workflow, you can use Scaden's function scaden example to generate example data and go through the whole pipeline.

First, make an example data directory and generate the example data:

mkdir example_data
scaden example --out example_data/

This generates the files "example_counts.txt", "example_celltypes.txt" and "example_bulk_data.txt" in the "example_data" directory. Next, you can generate training data:

scaden simulate --data example_data/ -n 100 --pattern "*_counts.txt

This generates 100 samples of training data in your current working directory. The file you need for your next step is called "data.h5ad". Now you need to perform the preprocessing using the training data and the bulk data file:

scaden process data.h5ad example_data/example_bulk_data.txt

As a result, you should now have a file called "processed.h5ad" in your directory. Now you can perform training. The following command performs training for 5000 steps per model and saves the trained weights to the "model" directory, which will be created:

scaden train processed.h5ad --steps 5000 --model_dir model

Finally, you can use the trained model to perform prediction:

scaden predict --model_dir model example_data/example_bulk_data.txt

Now you should have a file called "scaden_predictions.txt" in your working directory, which contains your estimated cell compositions.

1. System requirements

Scaden was developed and tested on Linux (Ubuntu 16.04 and 18.04). It was not tested on Windows or Mac, but should also be usable on these systems when installing with Pip or Bioconda. Scaden does not require any special hardware (e.g. GPU), however we recommend to have at least 16 GB of memory.

Scaden requires Python 3. All package dependencies should be handled automatically when installing with pip or conda.