Scientists at Duke just taught AI to shrink proteins — and it could make some of the world's most expensive medicines dramatically cheaper.
A $3.5 million gene therapy dose.
A single injection.
That's not a typo. That's Hemgenix, for haemophilia.
Zolgensma? $2.1 million. Casgevy? $2.2 million.
Families in India have crowdfunded for a single dose.
Gene therapy works by packing a healthy gene into a hollowed-out virus shell and shipping it into your cells.
But those shells have a strict size limit.
Many life-saving genes simply don't fit.
That's the whole bottleneck.
A team led by Duke's Rohit Singh built an AI system that can shrink, stretch, and rewrite proteins — like editing a sentence while keeping its meaning intact.
Published in Nature on July 29.
Here's the trick:
The team tested it across a 17-fold size range — haemoglobin at 147 amino acids, all the way up to mTOR at 2,549.
mTOR got trimmed by over 500 amino acids and still held its shape.
Then came the real test: fluorescent proteins.
70,000 AI-generated candidates, filtered down to 8, built in human cells.
Six glowed.
The smallest working ones were shorter than 96% of every fluorescent protein in the standard database — carrying over 40 coordinated edits and still functioning.
They glow dimly for now, not lab-bright yet.
And Raygun only protects what evolution optimized — not what human engineers bolted on later.
Still a work in progress.
If tomorrow's gene therapies can be miniaturized without breaking, they could fit into cheaper, better-understood delivery vehicles.
Smaller protein.
Smaller cost.
Bigger reach.
That's the real headline hiding behind the science.
That's all for now!
