EUGENe

EUGENe (End-to-end Utilitarian Genomics ENgine) is an open-source Python toolkit for building, training, and evaluating deep learning models that predict regulatory activity from DNA sequence.

Regulatory genomics has a fundamental challenge: we need to understand how DNA sequence encodes the instructions for when, where, and how genes are expressed. Deep learning has emerged as a powerful approach for learning these sequence-to-function mappings. EUGENe makes it easy to apply these methods to your own data.

What EUGENe does

• Sequence preprocessing — flexible handling of genomic sequences, variant effects, and data augmentation
• Model library — standard architectures (DeepSEA, Basset, Enformer-style) plus modular building blocks for custom models
• Training infrastructure — PyTorch Lightning-based training with logging, checkpointing, and reproducibility
• Evaluation — performance metrics, attribution analysis, and in-silico mutagenesis
• Interpretability — feature attribution (saliency, DeepLIFT, ISM) to understand what sequence features drive predictions

Links

• GitHub
• Documentation
• Publication — Nature Computational Science (2023)
• Preprint