RNA splicing, the process by which cells edit genetic instructions, relies on signals often located thousands of DNA bases away from the splice sites they influence. Existing AI models have struggled to detect these distant regulatory elements, limiting their accuracy.
A team of researchers has developed an AI model specifically designed to capture long-range DNA signals, significantly improving splice-site prediction. The work, reported in a recent study, addresses a major hurdle in understanding how DNA sequence variations affect gene expression.
Accurate splicing prediction is fundamental to human health, as errors in this process are linked to numerous genetic disorders. The new model offers a more reliable tool for identifying disease-causing mutations hidden in non-coding regions of the genome.
The findings open avenues for better diagnostic capabilities and targeted therapies. Further validation in clinical datasets will be necessary before the model can be deployed in medical settings, the researchers noted.
Experts caution that the model's performance on diverse populations and rare variants remains to be fully tested, highlighting the need for ongoing refinement.