Hello Bioinformatics lovers,

Today's newsletter is a little late. I have a friend's family visiting us.

Last year I asked an AI to write me an awk one-liner that trimmed the first 10 bases off every read in a FASTQ file. It gave me something in three seconds. The syntax was valid. It ran.

STAR refused the output. EXITING because of FATAL ERROR in reads input: quality string length is not equal to sequence length. The command had trimmed the sequence lines and left the quality lines alone, so every record held a 90-base sequence next to a 100-character quality string.

That error cost me four minutes, and it is the good outcome. STAR has a length check, it fired, it printed the offending read, and it told me to fix my file. Loud failures are a solved problem.

That is the shape of the whole problem. AI-written code fails, and the parser catches the first kind for free. What survives the parser is the second kind: valid syntax, valid file, wrong biology. The only detector for that is a person who knows what the answer should look like before it arrives.

Secoond story in R. Ask for a DESeq2 script and you get a clean one. Import counts, build the DESeqDataSet, run the test, plot the volcano. The code is correct. DESeq2 tests the model you hand it. If your samples came off the sequencer in two batches and you write design = ~ condition, the batch signal loads onto your condition term and returns as biology, with small p-values. The fix is design = ~ batch + condition.

DESeq2 will never suggest it. Your knowledge of how the samples were generated is the only thing that will.

Andrew Ng said at the end of 2025 that telling people to skip learning to code because AI will automate it will be remembered as some of the worst career advice ever given. His reasoning: as coding gets easier, more people should code. I will add the bioinformatics footnote. Our mistakes do not throw errors. They publish.

Learn Bash, R, and Python well enough to read what the AI wrote and ask why it chose each argument. That question is the job now.

Reply and tell me about a time AI handed you code that ran clean and was still wrong. I read every reply, and the best ones become future issues.

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!