June 2, 2026
Same proteins, different drama
The Unreasonable Redundancy of Nature's Protein Folds
AI may fold millions of proteins, but commenters say nature keeps reusing the same hits
TLDR: The big reveal: feeding AI more proteins may not add much fresh variety, because nature keeps reusing the same basic shapes. Commenters were split between joking that evolution is a copy-paste artist and wondering whether AI could finally discover the shapes biology never found.
Scientists are trying to use AI to design future drugs and enzymes, hoping that if they feed models enough examples from nature, the machines will get better at inventing useful new molecules. But this post delivered a surprisingly messy reality check: even though nature has produced billions of protein sequences, many of them seem to collapse into the same old structural shapes again and again. In plain English, biology may look huge, but commenters loved the idea that under the hood it’s basically running reruns.
And the community absolutely ran with that. The biggest crowd-pleaser was the accusation that evolution is basically a shameless copycat, with one commenter joking about “liberal protein plagiarism.” Another took the more thoughtful route, saying this is what happens when nature finds shapes that are stable, useful, and hard to break—so it just keeps reusing them like the world’s laziest but most successful designer. That sparked the spiciest underlying question in the thread: has evolution missed cool shapes that AI might discover first? That’s the kind of comment that turns a dry research post into sci-fi gossip.
Meanwhile, the comic relief was elite. One reader admitted the article crashed their browser, while another, despite confessing they barely understood any of it, praised the site for the radical act of simply loading all the text on one page. In a discussion about cutting-edge biology, the real bipartisan consensus may have been: please, just give us readable websites.
Key Points
- •The article says deep-learning biomolecular models such as AlphaFold3 have improved prediction of biomolecular interactions and enabled design of drug-like molecules.
- •It describes scaling model size, compute, and data as the main recipe for improving deep learning systems, including biomolecular models.
- •The article explains that AlphaFold3 turned large-scale protein sequence data into structural data by predicting 3D folds from natural sequences.
- •Ligo found that although natural protein sequences are extremely numerous, the underlying protein folds are much more redundant than sequence counts suggest.
- •The article argues that this fold redundancy limits how much new structural diversity can be gained by simply folding more natural sequences for enzyme-design training data.