DNABERT

Why we included this project

DNABERT brings the transformer approach that reshaped NLP to genome analysis. It converts raw DNA sequence into k-mer tokens and pre-trains a BERT-style encoder, which you can then fine-tune for tasks like promoter prediction, splice-site detection, and transcription factor binding site identification. The repo includes the model source code, usage examples, pre-trained checkpoints, and a visualization tool, so you can move from a genome file to a task-specific model without building the architecture yourself. Since it extends Hugging Face Transformers, the code will look familiar if you have worked in that ecosystem. One thing to know: the newer DNABERT-2, trained on multi-species genomes and more efficient, lives in a separate repository.