The Russian research institute AIRI has introduced an innovative tool called GENATATOR, based on neural networks, for high-precision gene mapping from DNA sequences. This development can find gene boundaries, determine transcript types, and reconstruct the internal structure of the genome.

How GENATATOR Works

The system consists of several AI models operating in stages. First, the neural network identifies probable transcript start and end points, then examines the regions between them, refines the structure, and filters out questionable predictions. The tool is designed to identify both protein-coding genes and long non-coding RNA genes.

Training and Testing

The models were trained on genes from humans and 38 other mammalian species, including walruses and elephants. During testing, GENATATOR demonstrated impressive results even on organisms not included in the training set: fruit flies, Arabidopsis thaliana flowers, and baker's yeast. In terms of accuracy, the solution significantly outperformed existing analogs.

Key Advantage

Unlike classical approaches, GENATATOR is not tied to searching for individual markers such as start and stop codons or splicing signals. Instead, the neural network memorizes and learns patterns in entire DNA sequences. This opens up the possibility of applying the tool to genomic assemblies of non-model organisms without detailed annotations.

Potential Applications

Theoretically, GENATATOR can be used to study evolutionary processes, discover new genes, and analyze ancient animals. In medicine, it is useful for disease prediction, creating personalized drugs, and biotechnological developments.

Availability and Limitations

The models are publicly available on Hugging Face, along with a web service and an open leaderboard for quality assessment. However, the local version only supports execution on a CUDA GPU with data output in float32 format. Execution on a central processor and lower-precision modes are not yet available.

Expert Opinion: GENATATOR is not just another bioinformatics tool but a breakthrough in genome analysis. The ability to work with non-model organisms without annotations significantly expands horizons for fundamental science and biotechnology. It is especially important that the technology is open to the community, which will accelerate its adoption and development.