Lightnews — Scholar-powered news

Ming Tommy Tang @tommytang.bsky.social · 20m

Investigating whether deep learning models for co-folding learn the physics of protein-ligand interactions www.nature.com/articles/s4...

Ming Tommy Tang @tommytang.bsky.social · 50m

I hope you've found this post helpful.

Follow me for more.

Subscribe to my FREE newsletter chatomics to learn bioinformatics divingintogeneticsandgenomics.ck.page/profile

Hi! I'm Tommy Tang

I am a bioinformatician/computational biologist with six years of wet lab experience and over 12 years of computation experience. I will help you to learn computational skills to tame astronomical data and derive insights. Check out the resources I offer below and sign up for my newsletter!

divingintogeneticsandgenomics.ck.page

Ming Tommy Tang @tommytang.bsky.social · 50m

10/
Key takeaways:
Login shell sets up your login environment

Interactive shell is for live terminal use

Non-login, non-interactive is how jobs run

Always source configs in scripts

Know your shell to debug faster

Learn your shell. Your scripts will thank you.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

9/
Another tip:
Use env or printenv to check your environment.
Use ps -f to see if you're in a login shell (-bash).
Test before you submit a 3-day job that fails in 2 seconds.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

8/
Solution:
At the top of your job script, add:
source ~/.bashrc

or
source ~/.bash_profile

Or define everything explicitly in the script.
Never assume the environment is set.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

7/
Example:
Your .bash_profile loads conda:

source ~/miniconda3/etc/profile.d/conda.sh

But your job crashes with command not found.
Why?
Because your script was run by a non-login shell.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

6/
Running a job on an HPC cluster?
SLURM calls a non-interactive, non-login shell by default.
Which means:
Your PATH, PYTHONPATH, R_LIBS_USER?
Might not be loaded unless explicitly sourced.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

5/
.bash_profile is read by login shells
.bashrc is read by interactive non-login shells
If you export paths only in .bashrc,
they may be missing when you login via SSH.
Boom—your pipeline crashes.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

4/
Why should bioinformaticians care?
Because .bashrc and .bash_profile behave differently.
And your environment might break if you don’t know which is loaded.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

3/
Can a shell be both login and interactive?
Yes!
If you SSH into a machine and start typing—
That’s both login and interactive.
But not all shells are both.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

2/
What’s an interactive shell?
A shell that reads and writes directly from a user’s terminal.
You’re typing into it.
You start one by typing:
bash

Or just open a new terminal tab.

1

Ming Tommy Tang @tommytang.bsky.social · 50m

1/
What’s a login shell?
It’s the shell that starts when you log in via SSH, a virtual console, or a terminal emulator set to launch a login shell.
You'll see it in ps as -bash (the hyphen means login shell).
It sets up your environment.

1 1

Ming Tommy Tang @tommytang.bsky.social · 50m

Bioinformatics is more about biology and code.
It's shells and sessions, exports and environments.
Ever been confused by “login shell” vs “interactive shell”?
This thread is for you. 🧵

1 1

Ming Tommy Tang @tommytang.bsky.social · 1h

I hope you've found this post helpful.

Follow me for more.

Subscribe to my FREE newsletter chatomics to learn bioinformatics divingintogeneticsandgenomics.ck.page/profile

Hi! I'm Tommy Tang

I am a bioinformatician/computational biologist with six years of wet lab experience and over 12 years of computation experience. I will help you to learn computational skills to tame astronomical data and derive insights. Check out the resources I offer below and sign up for my newsletter!

divingintogeneticsandgenomics.ck.page

Ming Tommy Tang @tommytang.bsky.social · 1h

14/
So next time someone asks:
“How long will it take?”
You can smile and say:
“It depends—how much truth are we ready to uncover?”

1

Ming Tommy Tang @tommytang.bsky.social · 1h

13/
Key takeaways:
“Bioinformatics” isn’t just one step

Routine plots are fast—insight takes time

Depth of analysis depends on the biological question

Interpretation evolves as the science unfolds

1

Ming Tommy Tang @tommytang.bsky.social · 1h

12/
And all of this?
It can take weeks.
Months.
Sometimes the entire length of a PhD (FOUR YEARS!).

1

Ming Tommy Tang @tommytang.bsky.social · 1h

11/
Then there’s visual storytelling:
Clear plots.
Accurate legends.
Clarity in PCA, UMAP, violin plots.
You’re not just analyzing. You’re communicating.

1

Ming Tommy Tang @tommytang.bsky.social · 1h

10/
You start with DESeq2.
Try edgeR.
Maybe limma-voom.
The results overlap, but differences remain.
That’s the art: building confidence across tools.

1

Ming Tommy Tang @tommytang.bsky.social · 1h

9/
You check for batch effects.
You wonder if a different normalization method changes the conclusion.
Spoiler: it might.

1 1

Ming Tommy Tang @tommytang.bsky.social · 1h

8/
The collaborator asks: “Can we validate this in another dataset?”
Now you’re off downloading GEO, harmonizing formats, rerunning pipelines.

1

Ming Tommy Tang @tommytang.bsky.social · 1h

7/
Then biology calls:
“Look at this gene—we know it's a target of the treatment. What’s it doing?”
Suddenly you're deep-diving into specific genes.

1

Ming Tommy Tang @tommytang.bsky.social · 1h

6/
You try clustering: maybe k-means or hierarchical.
You group genes with similar expression dynamics.
Each cluster tells a story—if you’re willing to listen.

1

Ming Tommy Tang @tommytang.bsky.social · 1h

5/
Which genes respond to Treatment1 but not Treatment2?
What’s synergistic in the Combo?
What’s additive vs unique?

1

Ming Tommy Tang @tommytang.bsky.social · 1h

4/
But you have 4 conditions:
Control, Treatment1, Treatment2, Combo.
Now the questions begin...

1