DeepVRegulome: DNABERT-based deep-learning framework for predicting the functional impact of short genomic variants on the human regulome

The introduction of DeepVRegulome marks a significant advancement in genomic research, particularly in understanding the functional impacts of non-coding short variants. By leveraging 700 DNABERT fine-tuned models, this framework effectively predicts and interprets functionally disruptive variants within the human regulome. Its application to the TCGA glioblastoma whole-genome sequencing dataset has revealed 572 splice-disrupting mutations and 9,837 mutations affecting transcription-factor binding sites, highlighting the framework's capability to prioritize clinically relevant mutations. Furthermore, survival analysis linked 1,352 mutations and 563 disrupted regulatory regions to patient outcomes, enabling better stratification of glioblastoma patients based on non-coding mutation signatures. This development not only enhances our understanding of genetic variations but also holds the potential to improve clinical decision-making and patient care in oncology.