Interactive regulatory variant-effect prediction using 464 fine-tuned DNABERT models
DeepVRegulome is a deep-learning framework for predicting the functional impact of short genomic variants in non-coding regulatory regions. The framework uses fine-tuned [DNABERT](https://github.com/jerryji1993/DNABERT) models trained on ENCODE ChIP-seq data to predict transcription-factor binding, histone-mark enrichment, splice-site activity, and allele-specific regulatory effects. This Hugging Face Space provides a lightweight interactive demonstration of the DeepVRegulome prediction framework. ## Model Coverage DeepVRegulome contains **464 fine-tuned DNABERT models**: - **458 transcription-factor binding models** - **4 histone-modification models** - **1 splice-acceptor model** - **1 splice-donor model** The transcription-factor and histone models were trained using regulatory regions derived from ENCODE ChIP-seq experiments. ## Features ### Binding Prediction Select a transcription-factor or histone-mark model and enter a DNA sequence to estimate its predicted binding or regulatory activity. The interface accepts a standard **301 bp DNA sequence**, although shorter or longer sequences may also be processed. ### Variant Effect Scoring Compare reference and alternate DNA sequences to estimate how a genomic variant changes predicted regulatory activity. The output includes: - Reference-allele probability - Alternate-allele probability - Change in predicted binding - Log-odds-ratio variant-effect score - Predicted gain or loss of regulatory activity ### Splice-Site Prediction (Added in PyPi package, Will be added in the Space) The splice-acceptor and splice-donor models can be used to evaluate sequence changes near exon–intron boundaries and identify variants that may alter predicted splice-site activity. ## Using the Demo 1. Select **Binding Prediction** or **Variant Effect Scoring**. 2. Choose a model from the model menu. 3. Paste the required DNA sequence or sequences. 4. Click **Predict**. 5. Review the predicted probabilities and variant-effect results. A random DNA-sequence generator is also available for quickly testing the interface. ## Python Package Install DeepVRegulome from PyPI: ```bash pip install deepvregulome ``` ## Links - **Paper**: [arXiv:2511.09026](https://arxiv.org/abs/2511.09026) - **Models**: [duttaprat/DeepVRegulome](https://huggingface.co/duttaprat/DeepVRegulome) - **Code**: [GitHub](https://github.com/DavuluriLab/DeepVRegulome) - **PyPI**: `pip install deepvregulome` - **Full Web App**: [deepvregulome.streamlit.app](https://deepvregulome.streamlit.app)