A group of scientists from the ARC Institute has developed fundamentally new model artificial intelligence called EVO, capable of analyzing and interpreting Biological sequences. Unlike the usual language models like Google Gemini or ChatGPT, the development was not studied on texts, but on the genetic material of millions of microorganisms.
The creators of the technology set themselves an ambitious task – to create a fundamental model for working with genomic data. EVO analyzes the sequences of DNA, RNA and proteins, just as language models process words and sentences. Moreover, each pair of DNA bases is perceived by the algorithm as a separate “word” in a huge biological text.
The training base includes information about 2.7 million prokaryotes and phages. Such a large -scale volume of data allowed the models not only to study existing sequences, but also to predict how small changes in the genetic code can affect the whole organism.
The creators of EVO emphasize the complexity of the task – even the simplest microbial genomes have incredible complexity. Despite this, the technologies managed to achieve a deep understanding of the genetic code, starting with the basic DNA elements and ending with whole genomes.
Technology operates simultaneously at several levels at once. It is taken into account both the multi -modality of the central dogma of molecular biology (the relationship of DNA, RNA and proteins), and the hierarchical nature of evolution – from individual molecules to entire organisms.
In practice, EVO generates realistic sequences with a whole genome length and even designs new biological systems. Laboratory validation of the synthetic systems of CRISPR and the transposons IS200/IS605, created using artificial intelligence, has already been carried out.
Another important achievement is the ability of EVO to create combinations of proteins and RNA that provide protection against viral infections. However, the technology is not yet perfect – some generated DNA sequences turned out to be non -functional, like a blurry photograph instead of a clear image.
The current version of EVO 1.0 and is not yet ready to work with the human genome. However, the very fact of the successful use of machine learning in the field of molecular biology opens up tremendous prospects for future research.