AI-Designed DNA Controls Genes in Healthy Mammalian Cells for the First Time: Study

In a landmark scientific breakthrough, researchers have successfully used generative artificial intelligence (AI) to design synthetic DNA molecules that can control gene expression in healthy mammalian cells—a feat never accomplished before. The study, conducted at the Centre for Genomic Regulation (CRG) in Barcelona, signifies a major leap forward in synthetic biology and gene therapy.

As a proof-of-concept, the research team directed the AI model to design synthetic DNA fragments capable of activating a specific gene coding for a fluorescent protein. The catch? It had to work in only select cells without disturbing other gene expression patterns. The AI rose to the challenge, generating unique DNA sequences from scratch. These were then synthesized chemically and introduced into mouse blood cells. Once inside, the sequences randomly integrated into the cells’ genomes—and remarkably, performed exactly as predicted.

"This is like writing software, but for biology," said Dr. Robert Fromel, the study’s first author. "It gives us entirely new ways of giving instructions to a cell and guiding how they develop and behave with unprecedented accuracy."

How It Works

The AI model functions much like a natural language model, but instead of processing words, it deciphers and generates DNA code—combinations of the four nucleotides: A, T, C, and G. Researchers can instruct the model to create fragments with highly specific functions. For example, they could ask it to “switch this gene on in stem cells destined to become red blood cells, but not in those becoming platelets.” The model then predicts the optimal sequence to produce that exact effect.

These synthetic enhancers are about 250 DNA letters long and can be delivered into cells via viral vectors. Because they are custom-designed, they can act as ultra-specific genetic switches—turning genes on or off only in the desired cell types.

Implications for Medicine and Biotechnology

The implications of this research are vast. For gene therapy, the technology could offer a powerful new way to boost or silence genes only in affected tissues—greatly increasing precision and reducing side effects. Unlike protein-based drugs, which often lack cell-type specificity, AI-designed DNA enhancers offer a new class of biological tools that could address the root causes of many diseases tied to faulty gene expression.

“Many human diseases are driven by the misfiring of genes in specific cells,” explained Dr. Lars Velten, corresponding author of the study. “Our goal was to crack the grammar rules of how enhancers work, so we can write entirely new sentences that cells will understand.”

While generative AI has already transformed protein design—speeding up the creation of novel enzymes and antibodies—this study marks a key milestone for its use in controlling DNA-level instructions. The challenge, however, lies in data. Training AI to understand enhancer sequences requires large volumes of high-quality genomic data, something historically difficult to obtain.

Still, the research lays the groundwork for a new era of generative biology, where AI and synthetic DNA merge to offer unprecedented control over life at the molecular level. The future could see tailored gene switches for personalized medicine, cell engineering, and even regenerative therapies that repair tissues by reprogramming cells at will.