Jupyter notebooks are a popular way to track the work performed in computational experiments the way a lab notebook tracks the work done in a wet lab setting. DXJupyterLab is an application provided by DNAnexus that allows you to perform computational experiments in the DNAnexus cloud using Jupyter notebooks. DXJupyterLab allows users on the DNAnexus platform to collaborate on notebooks and extends JupyterLab with options for directly accessing a DNAnexus project from the JupyterLab environment.
DXJupyterLab supports the use of Bioconductor and Bioconda, useful tools for bioinformatics analysis.
DXJupyterlab is a versatile application that can be used to:
Collaborate on exploratory analysis of data
Reproduce and fork work performed in computational analyses
Visualize and gain insights into data generated from biological experiments
Create figures and tables for scientific publications
Build and test algorithms directly in the cloud before creating DNAnexus apps and workflows
Test and train machine/deep learning models
Interactively run commands on a terminal
JupyterLab can be run on the DNAnexus platform in two different ways: a general-purpose JupyterLab application and a Spark cluster-enabled one, which can be used within the DNAnexus Apollo framework.
Both apps instantiate a JupyterLab server that allows for data analyses to be interactively performed in Jupyter notebooks on a DNAnexus worker.
The DXJupyterLab Spark Cluster app contains all the features found in the general-purpose DXJupyterLab along with access to a fully-managed, on-demand Spark cluster for big data processing and translational informatics.
A step-by-step guide on how to start with DXJupyterLab and create and edit Jupyter notebooks can be found in the Quickstart.
Creating a DXJupyterlab session requires the use of two different environments:
The DNAnexus project (accessible through the web platform and the CLI)
The worker execution environment
You have direct access to the project in which the application is run from the JupyterLab session. The project file browser (which lists folders, notebooks, and other files in the project) can be accessed from the DNAnexus tab in the left sidebar or from the terminal:
You can create, edit, and save notebooks directly in the project as well as duplicate, delete, or download them to your local machine. The project is selected when the DXJupyterLab app is started and cannot be subsequently changed.
The project file browser displays all subfolders, all Jupyter notebooks, and files in the project (limited to the 1,000 most recently modified files). In the Spark-enabled app databases in the project are also visible (limited to the 1,000 most recently modified databases). A list of all the objects in the project can be obtained programmatically or by using
dx ls. The file listing is refreshed every 10s and it is possible to enforce a refresh by clicking on the whirl arrow icon in the top right corner of the file browser.
NOTE: Jupyter is an application on another server, and not a DNAnexus product. Thus, any input data used by the notebook should first be downloaded to the execution environment from the DNAnexus project so it can be accessible from the JupyterLab notebook. Notebooks are executed in temporary execution containers and do not exist in the user’s local files. Any outputs from the notebook will also be generated in the temporary execution environment and should be uploaded to your DNAnexus project before the session ends.
When you open and run a notebook from the DNAnexus file browser the kernel corresponding to this notebook is started in this environment and the notebook code will be executed here.
The execution environment file browser can be accessed from the left sidebar (notice the folder icon at the top) or from the terminal:
Collaborators can work on notebooks in the project without the risk of overwriting each other's changes.
If a user has opened a specific notebook in a JupyterLab session, other users cannot open or edit the notebook. This is indicated by a red lock icon next to the notebook's name.
It is still possible to create a duplicate to see what changes are being saved in the locked notebook or to continue work on this "forked" version of the notebook. To copy a notebook, right-click on its name and select
Duplicate; after a few seconds, a notebook with the same name and a "copy" suffix should appear in the project.
Once the editing user closes the notebook, the lock will be released and anybody else with access to the project will be able to open it.
Whenever a notebook is saved in the project, it is uploaded to the platform as a new file that will replace the previous version, i.e. the file of the same name. The previous version is moved to the
.Notebook_archive folder with a timestamp suffix added to its name and its ID is saved in the
properties of the new file. Saving notebooks directly in the project ensures that your analyses won't be lost when the DXJupyterLab session ends.
DXJupyterLab sessions begin with a set duration, after which they will shut down automatically. The timeout clock is displayed in the footer on the right side and it can also be adjusted there (using the
Update duration button). Even if the DxJupyterLab webpage is closed, the termination will be executed at the set timestamp. Job lengths have an upper limit of 30 days, which cannot be extended.
A session can be terminated immediately from the top menu (
It is possible to save the current session environment and data and reload it later by creating a session snapshot (
A DXJupyterLab session is run in a Docker container, and a session snapshot file is a tarball generated by saving the Docker container state (with the
docker commit and
docker save commands). Any installed packages and files created locally are saved to a snapshot file, with the exception of directories
/mnt/, which are not included. This file is then uploaded to the project to
.Notebook_snapshots and can be passed as input the next time the app is started.
NOTE: If many large files are created locally, the resulting snapshots will take very long to save and load. In general, it is recommended not to snapshot more than 1 GB of locally saved data/packages and rely on downloading larger files as needed.
Viewing any other file types from your project, such as CSV, JSON, PDF files, images, scripts, etc. is convenient because JupyterLab will render them accordingly. For example, JSON files will be collapsible and easy to navigate and CSV files will be presented in the tabular format.
However, editing and saving any open files from the project other than IPython notebooks will result in an error.
The JupyterLab apps are run in a specific project, defined at start time, and this project cannot be subsequently changed. The job associated with the JupyterLab app has CONTRIBUTE access to the project in which it is run.
When running DXJupyterLab app, it is possible to view, but not update, other projects the user has access to. This enhanced scope is required to be able to read databases which may be located in different projects and are not cloneable.
It is possible to start new jobs with
dx run from a notebook or the Terminal. If the
billToof the project that the JupyterLab session is running in is not licensed to start detached executions, the started jobs will be the subjobs of the interactive JupyterLab session. In this case, the
--project argument to
dx runwill be ignored and the JupyterLab job's workspace is used to run the job, not the given project. Also, if the attached subjob fails or is terminated on the DNAnexus platform, the whole job tree will be terminated, including the interactive JupyterLab session.
The DXJupyterLab app is a Docker-based app that runs the JupyterLab server instance in a Docker container. The server runs on port 443. Because the app is an HTTPS app, you can bring up the JupyterLab environment in a web browser using the URL
job-xxxx is the ID of the job that runs the app. Only the user who launched the JupyterLab job has access to the JupyterLab environment. Other users will see a “403 Permission Forbidden” message under the JupyterLab session's URL.
NOTE: The JupyterLab server runs in the Miniconda Python 3.6 environment and Ubuntu:18.04 is the container's operating system.
In addition to
dx-toolkit, the JupyterLab server comes pre-installed with a number of popular data packages:
Kernels available for the notebooks:
CUDA libraries and drivers are installed on GPU-enabled instances.
Additional packages can easily be installed during a JupyterLab session. By creating a Docker container snapshot, users can then start subsequent sessions with the new packages pre-installed by providing the snapshot as input.
You can find more information on the features and benefits of JupyterLab in the official JupyterLab documentation.