Comprehensive Summary
The paper introduces a new large language model, GQ-DNABERT, which aims to improve G-quadruplex (GQ) mapping techniques and further understanding of the role of GQ confirmations in tissue-specific gene expression of both normal and cancer cells. The DNABERT model was trained using the highly-validated EndoQuad GQ database which allows for the prediction of GQ regions across the human genome. Model performance was validated through comparison with established experimental datasets, achieving high recall (0.9993) and precision (0.9977), indicating its reliability for genome-wide prediction. GQ-DNABERT also identified de novo GQs enriched in cis-regulatory elements (cCREs) and open chromatin regions (ATAC-seq peaks). Predicted GQ enhancer–promoter (pEP) pairs were found to be correlated with gene expression, with some acting as tissue-specific regulatory switches. In cancer cells, however, it was found that GQ pEP pairs were enriched in processes that turn on pathways essential to cell division and metabolism. Overall, GQ-DNABERT is presented as a reliable computational framework capable of predicting GQ regions in the whole human genome and advancing knowledge of transcriptional regulation at single-cell resolution.
Outcomes and Implications
The development of the GQ-DNABERT framework carries significant medical implications, as G-quadruplex (GQ) structures regulate gene expression, potentially influencing a range of new therapeutics. The improved understanding of how GQs influence transcriptional regulation helps reveal the mechanisms and pathways that drive oncogenic processes. This model also demonstrates potential for clinical research, as it allows for the identification of therapeutic targets to exploit GQ-faciliated regulation and to reprogram diseased cells, ultimately improving our understanding of the impact of GQ on cell fate.