A DNA designed with the help of artificial intelligence controls the behavior of mammalian cells for the first time.

A team of scientists in Barcelona has created an artificial intelligence system capable of designing DNA fragments that control the functioning of healthy mammalian cells for the first time.

The work represents a new twist on the use of machine learning systems in biomedicine. This technology is revolutionary, in the words of one of its main proponents, bioinformatician David Baker , who won the Nobel Prize in Chemistry for his work in this field last year. Until now, most of these applications have focused on producing proteins with custom-designed functions, in many cases creating molecules that did not exist in nature and that can function as vaccines, cancer treatments, or antidotes against poisons.

The new work, published in the specialized magazine Cell doesn't focus on protein design, but rather on the genetic code of DNA that contains the recipe for producing them. The human genome is a very long sequence of approximately 3 billion DNA letters (TCAGG, etc.). Although that instruction book has been read and some of its fundamental components, such as genes, are known, other parts of the code are still largely unknown, although they play a key role in determining how a human being is formed with all its differentiated tissue types, or how a tumor develops.

Lars Velten , a 37-year-old German biologist, has spent most of his career trying to understand the language of DNA. Specifically, the genetic elements that regulate gene function and determine, for example, whether a stem cell becomes a red blood cell capable of transporting oxygen throughout the body, or a white blood cell capable of seeking out and eliminating any threat. Scientists have focused on relatively small DNA fragments, about 250 DNA letters long, called enhancers, which are key to modulating the activity of genes crucial to the development and behavior of blood cells.

Over the past five years, the team has presented some 64,000 synthetic enhancers to an artificial intelligence system, which has learned the function of each one. It is the largest collection of these genetic building blocks ever assembled to understand the behavior of seven different types of blood cells, including red blood cells, several types of white blood cells , and blood stem cells. The enhancers work by binding to transcription factors, proteins that also modulate gene function. The researchers analyzed their exact combination with 38 transcription factors.

With all this data, the system was able to create novel enhancers that don't exist in nature. The researchers took these DNA fragments, introduced them into the genome of blood cells, and demonstrated that they are capable of turning on, off, or modulating the activity of the desired genes. A DNA word devised by an artificial intelligence system thus determines the behavior and fate of living cells.

"This is the first time something like this has been achieved in healthy cells, as until now research has focused on cancer cells, which are easier to handle," Velten emphasizes. The team has demonstrated the system using mouse blood cells, but they believe this is only the first step.

One possibility is that this same method could be used to control the behavior and fate of cells in other healthy tissues. Another more long-term application is to use this system in cancer cells, or even in other cells that already have dangerous genetic mutations that could lead to disease in the future. “There are certain mutations that accumulate with aging. I would be interested in developing genetic enhancers for these cells, because there is currently no good drug to combat them,” explains Velten.

Biochemist Susana Vázquez , a specialist in protein design using artificial intelligence at the Spanish National Cancer Research Centre (CNIO), who was not involved in the work, highlights its importance. “One of the most interesting aspects is how this extensive data collection has allowed us to train artificial intelligence algorithms capable of designing DNA sequences de novo, that is, from scratch,” the scientist says. “These sequences can induce specific cellular responses, which opens the door to a way of programming cellular behavior. I think this study perfectly reflects the transformative moment we are experiencing, where artificial intelligence is beginning to have a real impact in all areas of knowledge,” she adds.