Hello Bioinformatics lovers,

Tommy here.

I just finished interviewing candidates for one of the more unusual roles: half wet lab, half bioinformatics.

By the end, I had stopped listening for pipelines.

Fourteen years ago, I was a wet lab scientist teaching myself to analyze my own data. Many bench scientists are doing the same today.

The first months are steep, but the payoff is large. You get to answer your own questions, and your career options widen.

Curiosity and persistence carry you through.

Running DESeq2 is the entry ticket

Every candidate told me they could run a pipeline or a DESeq2 analysis. The strongest ones went further. They knew the biology deeply and told a story with their data that I believed.

AI now automates most preprocessing. The bottleneck has moved to interpretation, and interpretation needs someone who knows how the data was made. Wet lab experience lets you call out AI when it hands you nonsense.

The strand AI forgot

Here’s an example from my own work. I had Claude Sonnet search for estrogen receptor half sites (AGGTCA). It scanned only the forward strand.

A half site isn’t palindromic,

so motifs sitting on the reverse strand show up as TGACCT on the forward strand, and a one-strand scan misses all of them.

Claude Opus 5.5 scanned both strands without being told.

One run is an anecdote, and a prompt that says “scan both strands” fixes it. You have to know to say it.

A biologist knows a transcription factor can bind either strand. The model needed that knowledge supplied.

My current split: I use Opus to plan the analysis and Sonnet to write the scripts. Then I check everything, starting with QC. Batch effects and sequencing artifacts will sink an analysis before it starts.

Throwing raw data at AI and asking for results does not work.

Communication closes the loop

When AI produces the results, you still have to summarize them, and people have to trust you.

That means explaining how the analysis was done. You can ask AI to help draft that explanation too.

The candidates I want communicate well with two audiences: the model and their colleagues.

If you’re a bench scientist learning to analyze your own data, hit reply and tell me where you’re stuck. If you hire for hybrid roles, I’d like to hear what you screen for.

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/ Follow me there!

Stay awesome!