Artificial intelligence has fundamentally changed how scientists design proteins
Proteins are the machines of life — and designing new ones from scratch used to take years of painstaking work. AI has upended that in less than a decade.
Every process in a living cell is driven by proteins. They digest food, transmit signals, repair DNA damage, build cell walls, and destroy pathogens. Proteins are also the active components of most modern medicines. If you can design proteins at will, you can in principle engineer new biological functions — drugs that work more precisely, enzymes that degrade toxins faster, molecules that counteract the damage of aging.
But proteins are extraordinarily complex. An average human protein consists of hundreds of amino acids arranged in a precise sequence, which then fold into a specific three-dimensional shape. Small changes in that sequence can completely destroy function. Designing a working new protein by hand was the work of years — a combination of biochemical intuition, crystallographic research, and relentless experimentation.
What AI changed
The AI revolution in protein science began with AlphaFold, the DeepMind system that in 2021 essentially solved the decades-old problem of protein structure prediction — determining the three-dimensional shape of a protein from its amino acid sequence. That was a landmark. But a recent analysis in Science describes how AI is now answering the reverse question: not which structure corresponds to which sequence, but which sequence you need to obtain a desired structure and function. This is the ‘inverse problem’, and it is the heart of genuine protein design.
With tools like RFdiffusion and ProteinMPNN, researchers can now design proteins that do not exist in nature but that can perform specific tasks. In laboratory tests, a substantial fraction of AI-designed proteins proved functional — a success rate that would have been unthinkable a decade ago. The implications for medicine and biotechnology are significant: faster-developed therapies, cheaper production of biological drugs, and potentially new targets for diseases that remain untreatable today.
The connection to aging
For longevity science, this is particularly relevant. Many of the processes that drive aging are protein-dependent: the accumulation of misfolded proteins in the brain in Alzheimer’s and Parkinson’s disease, the decline of enzymatic repair processes, the dysfunction of mitochondria. If AI dramatically expands the toolkit for protein design, it opens new routes for interventions that were previously technically out of reach. How quickly those possibilities translate into clinical applications is another matter — but the acceleration is unmistakable.