Scientists built working proteins with one fewer building block — using AI
Life on Earth uses twenty amino acids to build every protein in every living thing. Researchers wondered whether all twenty are truly necessary — and used artificial intelligence to find out.
Twenty amino acids. That’s the alphabet of life — the molecular building blocks from which every protein in every known organism is assembled. Proteins do almost everything: they power cellular reactions, carry signals between tissues, defend against infection, and maintain the structural integrity of cells. The universality of this twenty-letter code is one of biology’s most striking features. But a study published in Science suggests it’s not quite as indispensable as it appears.
Using generative artificial intelligence — the same class of technology underlying large language models, but applied to protein sequences — researchers designed functional proteins that completely exclude one of the twenty amino acids. The proteins remained stable, properly folded, and functional. The excluded amino acid was cysteine, a building block involved in forming disulfide bonds — chemical links that give proteins structural rigidity and stability.
Why removing one building block matters for medicine
Cysteine is chemically reactive. It tends to form unwanted bonds with other molecules, which can cause problems during the manufacture of protein-based drugs and compromise their stability inside the body. Proteins engineered without cysteine are chemically simpler, potentially more stable, and easier to produce at scale. For the biopharmaceutical industry, that’s not a trivial advantage — it could translate into more reliable drug production and better-performing therapeutics.
For aging biology, the relevance runs deeper. Protein damage and misfolding are central features of biological aging and of many age-related diseases, from Alzheimer’s to Parkinson’s. The accumulation of damaged or misfolded proteins is one of the hallmarks of cellular aging. If therapeutic proteins can be designed with inherently better chemical stability — less susceptible to certain forms of damage — that could produce more robust biological drugs that perform better in aging tissues.
AI as a design engine for biology
The study also illustrates a broader shift in biology. Until recently, designing new functional proteins was painstaking work: scientists had to understand how amino acid sequences determine three-dimensional structure and then manually engineer variants. Generative AI can now produce and screen thousands of candidate proteins computationally before a single one is synthesized in a lab. That’s not an incremental speedup — it’s a qualitative change in what’s possible.
Whether a nineteen-amino acid alphabet will ever be widely deployed in drug development or synthetic biology remains uncertain. But the study demonstrates something important: the fundamental assumptions about what life requires to function are more flexible than biology’s universal code might suggest.