Hello Bioinformatics lovers,

Tommy here. When you read this, I am on my way to New York City for a conference. I am pre-scheduling it at 11:30pm.

I have been using Claude Code for a couple of month.

I love it and the capbility of coding has been increasing. I am very impressed by the opus 4.6 model.

However, just last week,

I almost sent a figure with wrong cluster labels to my collaborators made by Claude Code.

The R code ran. No errors. No warnings. The heatmap looked beautiful.

But cluster 2 showed upregulated genes in RNA-seq yet reduced chromatin accessibility in ATAC-seq.

That’s backwards — more transcription means more open chromatin, not less.

The code was correct. The cluster IDs weren’t consistent across heatmaps.

What happened

When you use row_km in ComplexHeatmap, the parameter cluster_row_slices defaults to TRUE.

This means the package re-orders row slices by hierarchical clustering on slice means after k-means splitting.

So “cluster 2” in your RNA-seq heatmap can contain completely different genes than “cluster 2” in your ATAC-seq heatmap. The labels match. The contents don’t.

The fix

Heatmap(mat, row_km = 4, cluster_row_slices = FALSE)

One parameter.

How I found it

Not from the code. From the biology. I know what drug-treated samples should look like. Claude doesn’t.

I asked Claude Code to debug the inconsistency.

Tip: use the “ultrathink” keyword for hard problems — it triggers deeper reasoning. (Note: this is a Claude Code feature. In the web chat, it won’t have the same effect.)

Claude found the root cause in minutes once I described the biological contradiction.

But it couldn’t have flagged the problem on its own.

The lesson

AI writes great code. It can’t judge if the output makes biological sense. That judgment comes from you — from years of looking at data and knowing what “right” looks like.

AI is a multiplier, not a replacement. Know your data. Know your biology. Let AI handle the syntax.

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, read my daily Linkedin post on AI and Bioinformatics https://www.linkedin.com/in/%F0%9F%8E%AF-ming-tommy-tang-40650014/recent-activity/all/​

Stay awesome!

PPS: