enformer-pytorch

Why we included this project

DeepMind's Enformer predicts gene expression from DNA sequence, and this repo is a PyTorch port that saves you from dealing with the original TensorFlow/Sonnet code. You feed in a sequence and get genome-wide tracks like gene expression and chromatin accessibility, with a pretrained model that matches the paper's reported correlation scores. It also supports fine-tuning for new tracks, conditioning on cell type or transcription factor context, and cross-attention over context embeddings. HuggingFace integration makes loading weights and shortening sequence length straightforward, and checkpointing helps fit the model on limited GPU memory. For researchers and bioinformatics engineers who want to reproduce Enformer results or build their own regulatory genomics models, this is a maintained, MIT-licensed starting point.