Protein Design
Protein design refers to the targeted development of new protein molecules with a predetermined function – today increasingly with the help of AI models that propose structures in minutes that used to take researchers years.
Proteins are the working molecules of life. They control metabolism, fight pathogens, and build tissue. Every protein consists of a long chain of building blocks, the amino acids, and folds into a specific three-dimensional shape. This shape determines the function. Protein design reverses this process: one first defines what a protein should do, and then specifically designs the amino acid chain that fulfills this task. In the past, this was done through years of trial and error in the lab. Today, AI models take over much of this work.
Reinventing proteins instead of just copying them
Nature has developed proteins over billions of years that solve specific tasks. However, these natural proteins are not sufficient for all the problems humans want to solve. Sometimes a protein is needed that blocks a specific cancer enzyme but does not attack healthy tissue. Such a protein does not necessarily exist in nature.
This is exactly where protein design comes in. It allows the construction of molecules with properties that do not occur in nature in this form. This is particularly valuable in medicine: designed proteins can act as highly precise drugs that only target where they should. But they are also in demand in industry, for example as enzymes that break down plastic or produce biofuel.
How AI calculates the amino acid chain
The classic approach was laborious: researchers modified known proteins step by step and tested each variant in the lab. This procedure is called rational design or directed evolution and can take years. AI models approach the problem differently. They were trained on millions of known protein structures and learned in the process which amino acid sequences lead to which shapes.
The best-known example is AlphaFold from the company DeepMind, which predicts the three-dimensional structure of a protein from its amino acid sequence. Newer models such as RFdiffusion or ProteinMPNN go a step further: they generate entirely new amino acid sequences for a desired target structure. The model proposes thousands of candidates within minutes. Lab scientists then test the most promising ones.
An important difference: the model only calculates on the computer. Whether the designed protein actually folds as predicted in the lab and then functions still has to be verified experimentally. AI dramatically shortens the search but does not completely replace the lab step.
Protein design in medicine, industry, and the news
Protein design mainly appears in the news in two contexts. First, with vaccines: some second-generation COVID-19 vaccines were based on artificially designed proteins that trigger a better immune response than natural variants. Second, with cancer drugs: researchers are developing designed proteins that specifically recognize and attack tumor cells without damaging healthy cells.
In industry, there are efforts to design enzymes that break down plastic waste. Natural enzymes can only do this slowly; designed variants are meant to do it faster and at higher temperatures. In 2023, David Baker, one of the pioneers of computational protein design, received the Nobel Prize in Chemistry – together with the developers of AlphaFold. This shows how central the field has become to science.
Anyone reading tech or biotech news frequently encounters protein design in connection with startups that want to develop AI-generated molecules directly as drugs. The term “AI drug design” usually means exactly that: not chemical active ingredients, but designed proteins that intervene in the body like custom-made tools.