DNABERT is a transformer model trained on DNA sequences. It uses BERT-style masked language modeling on large genomic sequence collections to learn statistical patterns associated with gene regulation, transcription factor binding, and sequence function. section 13.2.1
In drug discovery, genomic modeling can support target identification and validation. DNABERT applications include predicting the effects of non-coding variants on gene regulation, identifying regulatory elements such as enhancers and silencers, and designing gene therapy vectors for optimized expression. section 13.2.1
DNABERT is trained on DNA, not RNA. For RNA-focused tasks, generally use a model trained or fine-tuned on RNA instead. section 13.2.1
