Running a Reproducible Analysis
Project Management, Day 1
8.1 Reproducible science… in-silico??
- Bioinformaticians are people too
- We need to make sure our research is well documented and reproducible just like bench scientists
- Projects can get complex, messy, and very computationally demanding
8.2 How can computational projects get derailed?
It turns out that computational biologists need to be careful with how they manage their code and data. Leaving everything on your personal/lab computer comes with a lot of risk.
You can reduce the risk of a mishap by housing data on UNC’s cloud computing service, Longleaf, and putting your code on Github. Both of these provide you with backups that can be accessed from anywhere with the internet.
Don’t think it’s worth it? Here are some moments that made UNC researchers wished they had used these tools:
“…there was a time where my computer just stopped letting me log in and needed to be wiped so if I wasn’t using Longleaf I would have lost everything. I did lose a nice powerpoint.”
“In undergrad I was using local storage only on the desktop in my advisor’s office. There was some big failure with IT one day (tbh I still don’t know what happened) and I lost all my code”
“Our collaborator lost the hard drive with the raw RNAseq data, dooming my first 1st author publication. His collaborator saved the day with a backup he had on Longleaf”
“An undergrad left their code in /pine
* when they went home over the summer and it got deleted, so they had to re-write their code from scratch (which delayed the project as a whole)”
“Only having GitHub as a memory of projects I did in grad school, being able to search and find bits of code so I don’t have to rewrite them.”
“People emailing 2 years after a paper is published asking about obscure details of simulation etc.”
*pine
was UNC’s temporary data storage space
8.3 Good computational practices 101
Computational projects ought to be approached with the same expectations of rigor and reproducibility expected of a bench project. This means that the work needs to be well documented, things need to be properly stored, and everything should be organized clearly enough for someone to reproduce it.
Thankfully, we’re not the first researchers to run into these problems. A whole suite of tools and services exist to manage these issues:
Documenting everything
Storing data & getting resources
Keeping your R project organized
git
/GitHub
Longleaf
RStudio Projects
8.4 Suffering from manual version control? git
can help.
What is version control exactly? At its core, it’s a way of keeping track of the changes made to files. Before now, you’ve probably used a system like this:
paper_draft1.doc
paper_draft2.doc
paper_reviewed_by_john.doc
paper_draft3_comments_incorporated.doc
paper_final_draft.doc
paper_final_reviewed.doc
paper_final_submission.doc
paper_final_submission_revised.doc
paper_final_submission_revised_v2.doc paper_published_version.doc
With git
, you can update a file while keeping a detailed log of the changes.
* 9a2b3c4 - Add published version of the paper (2024-04-29)
* 8f7e6d5 - Revise submission after additional feedback, version 2 (2024-04-25)
* 7d6c5b4 - Update submission based on post-submission feedback (2024-04-20)
* 6c5b4a3 - Prepare final version for submission (2024-04-15)
* 5b4a392 - Finalize draft after thorough review (2024-04-10)
* 4a39881 - Incorporate feedback from final review (2024-04-05)
* 3928717 - Update draft, incorporate feedback from John (2024-04-01)
* 2871606 - Add second draft of the paper (2024-03-28)
* 1760505 - Initial draft of the paper (2024-03-25)
8.4.1 Go on, git
!
git
is version control system used to record changes to files. GitHub uses git
to help users host/review code and manage projects
git
/GitHub matter because they:
Track every version of every script
Publicly document your work
Allow for new versions of projects to
branch
Make it easy to collaborate
8.5 Longleaf: The darling of UNC bioinformaticians
Longleaf is UNC’s high-performance computing cluster (HPC). It’s basically a ton of computers/storage. Its accessible from anywhere with internet and offers a lot of storage. Labs typically start with 40 TB, users get 10 TB. Also you’ve been using it this whole time! RStudio OnDemand is hosted by Longleaf.
There are a ton of reasons to use LL:
Many scripts can be run at once, with your computer off
It has A LOT more resources than a typical computer
Easy to share files!
Dedicated technical support via ITS
8.6 Connecting Longleaf and Github
Github and Longleaf each can be daunting to novice programmers, so lets walk through how to set them up together.
The setup is going to amount to three general steps:
- Introduce our Github and Longleaf accounts to each other with something called a
SSH
key- SSH keys (secure shell) are encrypted passwords that link two computer systems
- Make our first repository (project) on Github
- Learn how to get scripts from Github to Longleaf and update changes we made on Longleaf back to Github
Before we can start that, we’re going to need to know just a tad about terminals and Bash. These topics could be a whole course onto itself, but in a nutshell you can think of them like this:
Terminals, also called command lines, are text-based software for interacting with your computer. RStudio has a built-in terminal, on the tab next to “Console”.
Bash is a type of computer language that understands and carries out the instructions you type in the terminal, usually called “shell scripting”. It’s very common on Linux and Mac computers. Longleaf uses Linux and working on Longleaf means using a bit of Bash.
This course isn’t meant to teach you shell scripting, and you aren’t expected to fully understand some of the Bash command we’ll run. If you’d like to learn more, please refer to the following cheat sheets on common bash commands and scripting in bash as helpful resources!
8.7 Linking via SSH
Let’s start by getting the terminal open. In the top left, click View -> Move Focus To Terminal. It should’ve opened in the panel that also contains tabs for “Console” and “Background Jobs”.
Next, let’s find the SSH key associated with your Longleaf account. We’ll run the following bash command in the terminal that we just opened (be aware that terminals are notoriously finicky with copy and paste): cat ~/.ssh/id_rsa.pub
Do NOT forget to add the .pub
of the above extension. RSA keys come in pairs, a private and a public version. The private key (i.e. the file that does not have the .pub
extension) should NEVER be shared. Sharing your private key will allow malicious actors to interact with services that cache your public key as if they were you!
This (hopefully) has copied your public SSH key to your clipboard!
Now, let’s go over to Github and set up the key.
Go to profile settings on github and select the “ssh key” section
Add new key
Name the key (should remind you that this is the key for Longleaf)
Paste the copied ssh key from the
cat
step aboveCreate!
With that done, we need to log into Github on Longleaf:
Go back to the RStudio terminal
git config --global user.name "your-github-username"
- Keep the quotes around your username!
git config --global user.email your.email.linkedwithgithub
- No quotes this time!
8.8 Making a repository
Great, we’ve connected Longleaf and Github (a herculean task for a beginner programmer!). What we’ll want to do next is make a repository (repo), which you could think of as a self-contained project folder. Let’s go back to Github:
In the “Repositories” tab of your profile, click “New”
Give it a name, maybe “example_repo”
Click the “Add a README file box”
Add a `.gitignore` with an R template
Create the repo!
We’ll explain the signifance of the README and .gitignore steps a bit later. For now, lets go over to our new repo on our profile. To get it onto Longleaf, we can:
Click on the green “Code” box
Click on “SSH”
Copy the SSH right below that! It should end in `.git`
Lets create a new directory to house our experimental projects. In the terminal window of RStudio, do the following:
cd ~ #set working directory to home directory mkdir learning-R #create a new directory "learning-R" cd learning-R #set "learning-R" as our working directory
Git Project OrganizationGit repositories may be cloned anywhere to your file system that makes sense to you. The above is a simple example of project organization in which we may create more repositories under the
learning-R
directory. Ultimately the choice is yours on how you wish to organize your git projects.Now we can clone our new github repository locally!
# add what you copied in step 3 behind "clone". # Your command may look something like this git clone git@github.com:<username>/<reponame>
8.9 RProjects
In this section, we will be discussing recommendations for organization. From Section 8.8, we can see that our project is fairly empty
.
├── .git/
├── .gitignore
└── README.md
It is up to use to bring some organization to our project!
The hidden directory .git/
and file .gitignore
will be covered in the next Section 8.10.
Organization can greatly improve the experience of coding. It is a way for us to show our future selves some kindness as well as anyone who may maintain our work in the future.
Below is an example of how we may setup directories for our project:
.
├── .git/
├── .gitignore
├── data/
├── outputs/
├── R/
├── scripts/
└── README.md
Here, we have created 4 new directories: data/
, outputs/
, R/
and scripts/
. These directory names imply something to the viewer about their contents and provide quick navigation.
data/
and outputs/
are perhaps the most self explanatory. We will reserve data
for files we may read in for our analysis, and outputs
as a place for us to store our results.
R/
and scripts/
may be a bit nuanced. Perhaps the project author will keep helpful reusable R code in the R/
directory, while the scripts/
directory may be analysis workflows that are called from the command line.
A project may, or may not, require this level of granularity. You may choose different directory names as well. However we want to maintain some level of interpretation and do not want to contradict expectations with the files within. Critically thinking about your project structure will ultimately save you time when you return to the project.
Here are some general guidelines to follow:
Organization Do’s
- Place files in some sort of relative structure
- Use descriptive file and directory names
- Format dates with
YYYY-MM-DD
Organization Do Not’s
- Place all files at the top level of the repository
- Stratifying files between too many directories
- Using spaces in file names!
8.9.1 README! … Please? 😢
As the name implies, this file is intended to be read by anyone who happens upon your repository.Think of this as extra documentation that you, as the project owner, may use to communicate aspects of the repository. Include helpful information about your project such as:
- A brief description of the project
- What is the scope of the project
- What problem does it solve?
- Intended use cases!
- Examples of what it should not be used for!
- How to get started or use the project
- How to contribute or report a bug
- Project Status: active development, stable, or abandoned
If your repository is being viewed on github.com then top level README.md
is displayed like the landing page.
You may also place README.md
within sub directories as well, allowing you to communicate more specific information within.
Here are examples of what you may include in a data/README.md
file
explanation of what data was collected and how it was processed.
documentation of the columns of a
.csv
file. What the acronyms and data labels mean and how they should be interpreted.
8.10 Using git
.git/
is a hidden directory whose main function is to track yourgit ...
commands. It is generally wise to ignore this directory..gitignore
uses rules to exclude files by name pattern or location- Don’t upload data or large files!
Changes need to be staged with
git add
git add .
adds all files not excluded by the .gitignore in the directorygit add -i
opens an interactive adding session
Commit staged changes with a note to your future self
git commit -m "Hi future me, this is what I changed"
note: Usingvim
If you were to write
git commit
and omit the-m "your message here"
portion,git
will force you to write a message by placing you into an interactive prompt. The default text editor is usuallyvim
which may be tricky to navigate.To enter edit mode, press
i
on your keyboard, and then you may begin writing your message. When you have finished, pressESC
to exit your current command mode. To exit vim, press:wq
which informs vim to write your changes and quit.
Commits are pushed to branches
- For the main branch, use
git push origin main
- For the main branch, use