scVI-container

This repository contains files for creating and testing containers to run code in the https://github.com/gvrocha/scVI-reproducibility repository. That repository was forked from https://github.com/romain-lopez/scVI-reproducibility and contains modifications to make it run smoothly with the container created herein.

The resulting containers are currently hosted at https://hub.docker.com/repository/docker/gvrocha/scvi

Description of the containers in this repo

Different versions of containers are made available in this repository for different purposes.

• A container based on TF1.15.2 for use with the scVI-reproducibility repo (https://github.com/romain-lopez/scVI-reproducibility);
• A container based on Pytorch for use with the scVI repo (https://github.com/YosefLab/scVI);
• A container based on TF2.0.1 that can be used to run current TensorFlow tutorials (e.g. )

Using the containers with Singularity

• To build a local Singularity image named scvi.py3-tf1.15.2.simg with the TF1.15.2 version:

singularity build scvi.py3-tf1.15.2.simg docker://gvrocha/scvi:py3-tf1.15.2

• To build a local Singularity image named scvi.py3-tf2.0.1.simg with the TF2.0.1 version:

singularity build scvi.py3-tf2.0.1.simg docker://gvrocha/scvi:py3-tf2.0.1

• To build a local Singularity image named scvi.py3-pytorch.simg with the Pytorch version:

singularity build scvi.py3-pytorch.simg docker://gvrocha/scvi:py3-pytorch

To execute within the Singularity container, two recommendations are made:

• use the --containall so the container is kept separate from the local environment,
• create a local directory containing this, the scVI-reproducibility and the scVI repos and link it to the /scVI directory in the container using the -B option.

Assuming these are followed and that the repos are stored in the /home/myusername/my.scvi.directory, the call to run the Singularity container is

singularity run --containall -B /home/myusername/my.scvi.directory:/scVI scvi_py3-tf1.15.2.simg

Using the containers with Docker

• To run a Docker container with the TensorFlow 1.15.2 version and linking the home/myusername/my.scvi.directory folder in the host to the /scVI folder in the container, call:

docker container run -it -v /home/myusername/my.scvi.directory:/scVI  gvrocha/scvi:py3-tf1.15.2 /bin/bash

• To run a Docker container with the TensorFlow 2.0.1 version and linking the home/myusername/my.scvi.directory folder in the host to the /scVI folder in the container, call:

docker container run -it -v /home/myusername/my.scvi.directory:/scVI  gvrocha/scvi:py3-tf2.0.1 /bin/bash

• To run a Docker container with the Pytorch version and linking the home/myusername/my.scvi.directory folder in the host to the /scVI folder in the container, call:

docker container run -it -v /home/myusername/my.scvi.directory:/scVI  gvrocha/scvi:py3-pytorch /bin/bash

Testing the image

After entering an instance of the image, the following tests can be performed.

Load tensor flow within python

For the tensor flow containers, run python within the container and call

import tensorflow as tf
print(tf.__version__)

Load libraries within R

For all containers, open an instance of R in the container and run

library(rmarkdown)
library(zinbwave)

Create an example report using Rmarkdown

In the examples/tests in this section, you may be prompted to install Miniconda. If that happens, just say no; the examples should still work without it.

Assuming that the folder containing the scVI-container repo is mounted at /scVI as recommended above, go to R and run

rmarkdown::render("/scVI/scVI-container/code/tensorflow_test.Rmd")

More detailed test for TF2.0.1

Under the same assumptions, you can also use the scvi:py3-tf2.0.1 container to generate a report that reproduces the TensorFlow tutorial within an Rmarkdown document by calling, from within R:

rmarkdown::render("/scVI/scVI-container/py3-tf2.0.1/code/tensorflow_example.Rmd")

Test for PyTorch/scVI

Within the scvi:py3-pytorch container, start R and, assuming the scVI container repo is mapper within /scVI, run

rmarkdown::render("/scVI/scVI-container/py3-pytorch/code/pytorch_test.Rmd")