EvoAug-TF

EvoAug-TF implements evolution-inspired data augmentation within TensorFlow to generate synthetic genomic sequences for training deep neural networks that predict molecular functions of non-coding genomic regions.

Key Features:

• Evolution-Inspired Augmentation: Generates synthetic training sequences using evolution-inspired transformations to augment functional genomics datasets and improve model generalization.
• TensorFlow Integration: Provides an implementation compatible with TensorFlow-based genomic deep learning architectures.
• Performance Benchmarking: Includes systematic benchmarking that demonstrates performance comparable to the original EvoAug package.
• Reproducibility Scripts: Supplies scripts to reproduce the reported benchmarking results.
• Support for Interpretability: Augmentation is intended to facilitate development of interpretable models and attribution analysis for genomic regulatory mechanisms.

Scientific Applications:

• Genomic DNN Training: Augments limited functional genomics datasets to enable training of deep neural networks for predicting molecular functions of non-coding regions.
• Model Generalization: Improves model generalization across diverse biological contexts by increasing training data diversity.
• Attribution and Regulatory Analysis: Supports attribution analysis to aid interpretation of learned sequence features and regulatory mechanisms.

Methodology:

Implements evolution-inspired data augmentation to generate synthetic training sequences within TensorFlow, performs systematic benchmarking against EvoAug, and provides scripts to reproduce benchmarking results.