AIRI has launched GENATATOR, a neural network tool for mapping genes, significantly improving accuracy in gene identification.
The Russian research institute AIRI has launched GENATATOR, a novel neural network tool designed for gene mapping using DNA sequences. According to a press release, this innovative system can identify gene boundaries, classify transcript types, and reconstruct their internal structures.
GENATATOR operates through a multi-stage process. It initially locates potential start and end points of transcripts, analyzes the intervening regions, refines the structural data, and eliminates unreliable predictions.
This advanced tool focuses on identifying both protein-coding genes and long non-coding RNA genes. Its models were trained using genetic data from humans and 38 other mammalian species, including walruses and elephants. Impressively, GENATATOR also demonstrated strong performance when tested on various other organisms such as fruit flies, Talya's Resuhovidka flower, and baker's yeast.
Notably, GENATATOR has surpassed existing models in accuracy. The AIRI team highlighted that it differs from traditional methods by not relying on specific coding gene markers such as start and stop codons or splicing signals. Instead, it learns patterns from DNA sequences in a comprehensive manner.
This capability enables the application of GENATATOR to genomic assemblies of non-model organisms that lack detailed annotations. Theoretically, it holds potential for studying evolutionary processes, discovering new genes, and examining ancient species. In the medical field, it could aid in disease prediction, development of personalized treatments, and biotechnological advancements.
Currently, the models are available on Hugging Face, along with a web service and an open leaderboard for evaluating quality. However, the local version supports only CUDA-compatible GPUs and outputs in float32 format, with no current options for CPU execution or lower precision modes.
In a related development, leaders from OpenAI, Anthropic, Google DeepMind, and Microsoft AI recently signed an open letter urging legislative measures for mandatory checks on clients and suppliers of synthetic DNA and RNA.