The model can process a DNA sequence of alternating nucleotides six times longer than DNABERT (a similar American model). The developers say, “the model was tested on one of the genetics problems: the prediction of sequences to “switch genes on” and has already shown results that are superior to those achieved using DNABERT”.

GENA_LM will allow to learn more about disease occurrence and the formation of malignant cells in the human body. AIRI plans to try using transformer architectures with memory, which will improve the model accuracy.

Original in Russian