Why aren't we using LLM's to chemically redesign old drugs?

[techteach00]

I was fooling around with Cisplatin, a chemo drug that made testicular cancer curable. Then I was like, I wonder how Chat GPT would improve this drug. It tried. Is this a good idea?

"The best chemical improvement is unlikely to be “a stronger cisplatin.” It should be a less reactive circulating molecule that becomes ordinary cisplatin only after tumor uptake. My preferred candidate is therefore: A neutral, asymmetric cisplatin-based Pt(IV) prodrug with one stabilized albumin-binding axial ligand and one reduction-released glutathione-synthesis inhibitor, with the axial electronics deliberately tuned to prevent premature blood reduction."

[ksaj]

I would go one step further with "Name and describe a few candidates that conform to this recommendation" and watch it enthusiastically hallucinate candidates.

I believe it'll get there one day. But it's pretty clear we're still at "incredible toy" stage.

[rationalist]

I imagine people are, they're just doing it in their research labs under confidentiality.

[verdverm]

LLMs are not the right type of Ai, look to AlphaFold and the mega dataset of protein folds they have made available for science. Things are happening, it will take many decades for all the insights and new ways of doing medicine to be mined and turned into effective treatments.

ChatGPT (llm) is definitely not the way, it is a sycophant and will tell you anything is a good idea.