Hello Bioinformatics lovers,

Tommy here. We will talk about why biological background is important for Bioinformatics analysis.

A single-cell dataset came back looking wrong to me. I put it down to noise and moved on.

The wet lab biologist who ran the experiment looked at the same plot and said the cells were under stress at that point. The pattern stopped being noise and became a result.

No parameter sweep would have gotten me there. The information was not in the count matrix. It was in the head of the person who handled the samples.

Analysis is an experiment

You start with a hypothesis. You design a dry lab test. You read the result, and when it comes back empty you refine the question and run it again.

That loop is what a bench scientist does, minus the pipettes. Treating a pipeline run as a finished product instead of one iteration is where most analyses go wrong.

Most of what I know about interpreting single-cell data I learned in conversations rather than from documentation.

A collaborator says one sentence about how the samples were handled, and a confusing result resolves.

Stress signatures cut both ways

van den Brink and colleagues showed in 2017 that tissue dissociation induces gene expression changes that can produce an artificial subpopulation within a cell type.

O'Flanagan and colleagues followed with a core set of 512 heat shock and stress response genes, including FOS and JUN, induced by collagenase digestion at 37°C and reduced by dissociating with a cold active protease at 6°C.

A stress cluster can be real biology or a handling artifact. Your count matrix cannot tell you which. The person who did the dissociation usually can.

Fill your own gaps, then verify

If your biology background has holes, LLMs close them faster than a literature search.

Ask what a pathway does, why a marker appears in this tissue, what is already known in this context.

Then check every reference before you cite it.

These models produce plausible-looking citations for papers that do not exist, and that failure mode targets exactly the people who cannot yet spot a wrong answer.

Running Seurat or DESeq2 is the easy part. Deciding whether the output means anything is the job, and that judgment comes from biology.

Reply and tell me about a time a collaborator's offhand comment changed how you read your own data. I read every response.

If you liked today's newsletter, foward it to someone else https://divingintogeneticsandgenomics.kit.com/profile. Sharing is caring :)

Happy Learning!

Tommy aka crazyhottommy

PS:

If you want to learn Bioinformatics, there are four ways that I can help:

1. My free YouTube Chatomics channel, make sure you subscribe to it.
2. I have many resources collected on my github here.
3. I have been writing blog posts for over 10 years https://divingintogeneticsandgenomics.com/​
4. Lastly, I post daily on Linkedin https://www.linkedin.com/in/%F0%9F%8E%AF-ming-tommy-tang-40650014/recent-activity/all/​

Stay awesome!