Artificial intelligence (AI) is increasingly entering the realm of biology, and companies like Living Models are trying to make a breakthrough that could change the way we understand life at the molecular level.
Instead of text and images, the new AI models are trained on DNA and RNA sequences. The idea is to view the genetic code as an information system, i.e. a kind of “programming language” of nature. As explained by Living Models, DNA encodes RNA, RNA encodes proteins, and they determine the characteristics of the organism.
DNA as a programming language powered by AI
This approach represents a big shift compared to previous methods. Traditional models mostly worked on the basis of correlations – linking certain genetic markers with traits, but without understanding the causes.
As stated in the analysis, the problem is no longer in data collection, but in their interpretation and generalization. Existing tools often fail to predict how the new genetic makeup will behave under different conditions.
Transformer models change the rules of the game
The solution comes from the same architecture that started the revolution in language processing – transformer models. Instead of “memorizing” patterns, they learn the structure and relationships within sequences.
In biology, this means recognizing regulatory patterns, interactions between distant parts of the genome, and functional connections that were previously unseen. The goal is to create a model that can “read” the genome, and not just analyze it superficially.
Why do plants come first?
The focus on plants is not accidental. There are several key reasons: data availability, less regulatory constraints and faster experimental cycles.
Unlike human genetics, where access and regulation are a major challenge, plant data is generally publicly available. Also, the results of experiments can be obtained in one season, instead of years or decades.
In addition, climate change creates an urgent need to develop more resistant crops, which makes this area an ideal field for the application of AI models.
In practice, such models can help breeders more quickly identify genetic variants that increase resistance to drought or heat. Instead of testing thousands of combinations, the focus is narrowed down to a smaller number of quality candidates.
It is important to emphasize that these systems do not make final decisions, but serve as a tool for hypothesis generation. The ultimate verification is still done by scientists through experiments.
Biology is increasingly turning into an information problem, and AI models trained on DNA could become a key tool for understanding and shaping life in the future, reports TechRadar Pro.