Hello Bioinformatics lovers,

Tommy here. Today's newsletter is Sponsored by Apodex

===============================

Four years ago I spent an afternoon hand-pulling metadata out of a GEO scRNA-seq dataset. I wrote a whole blog post about how annoying it was.

Last week I gave Apodex 1.1 the same job. One prompt: read this paper, find the supplementary file and the GEO dataset, give me a metadata CSV with responder status.

Supplementary Table S1 has no R/NR column. It derived the labels from the paper's own definition (CR/PR → R, SD/PD → NR), then spot-checked P007, P010, P012 and P019 to show the mapping held. It built the Seurat-ready cell-level table and reran the baseline CXCL13+ T cell comparison between responders and non-responders.

It reproduced the direction reported in the paper, though not at patient-level significance, and gave me all the code. I could trace every reasoning step.

Seeing the work matters more than getting the answer.

The agent framework is open source: ​

star the github repo​ ​

Download Huggingface weights Apodex 1.1 mini 35B weights

================================

Now, let's get into today's newsletter.

A wet lab scientist stops by your desk and says, "We think the drug causes isoform switching."

You nod, pull up your results, and realize your pipeline only ever counted reads at the gene level.

Every number you were about to show them answers a question they never asked.

I have sat on both sides of that moment. Fourteen years in, my fix has stayed the same: talk to the person who made the data before you touch it.

Where the friction starts

The dataset in your inbox took months, sometimes years, to produce.

The wet lab scientist lived through contamination, low yields, failed PCRs, and cell lines that refused to cooperate.

Then they handed it to you.

Your side carries its own pain. A bad FASTQ file. Poor replicates. Missing metadata. You cannot analyze what nobody measured correctly.

Frustration flows in both directions, and one honest conversation dissolves most of it.

Ask about biology first

I meet collaborators in person whenever I can. Zoom works, but sitting across a bench from someone builds trust faster.

Three questions open the conversation:

• What is the biological question?
• Which result would you care about most?
• What would change your next experiment?

The answers shape every downstream choice: experimental design, filtering thresholds, which annotation you use.

Nobody makes those calls well in isolation.

One example show why. Isoform switching needs transcript-level quantification, for instance salmon or kallisto output fed into tximport, or a dedicated package like IsoformSwitchAnalyzeR.

A gene-level count matrix from featureCounts cannot see it, however you tune the DESeq2 model.

Running differential expression and emailing a volcano plot counts as customer service.

Collaboration begins when you understand why they ran the experiment.

Respect runs both ways

Your collaborator may never learn that you spent hours aligning reads from low-quality libraries or fixing a broken annotation file.

Tell them. Then ask about the failed experiments you never heard about. You are solving the same puzzle from opposite ends.

Three habits I keep:

• Meet often, in person when you can.
• Ask about biology before analysis.
• Your best scripts come from your best conversations.

I wrote more about this in Hidden skills beyond programming for computational biology.

Hit reply and tell me about one conversation that changed how you analyzed a dataset. I read every response.

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/ make sure you follow along!​

Stay awesome!