EvoAug-TF: extending evolution-inspired data augmentations for genomic deep learning to TensorFlow.

Area of Science:

• Genomics
• Computational Biology
• Machine Learning

Background:

• Deep neural networks (DNNs) are crucial for predicting molecular functions of the non-coding genome.
• Limited experimental data in functional genomics hinders DNN training.
• Existing methods like EvoAug improve DNNs but are limited to PyTorch.

Purpose of the Study:

• To extend EvoAug's evolution-inspired data augmentation capabilities to TensorFlow-based deep learning models.
• To enable broader application of advanced DNN training techniques in genomic research.

Main Methods:

• Development of the EvoAug-TF package, adapting EvoAug for TensorFlow.
• Systematic benchmarking to compare EvoAug-TF performance against the original EvoAug.

Main Results:

• EvoAug-TF successfully extends EvoAug to TensorFlow, supporting a wider range of genomic DNNs.
• Performance benchmarks demonstrate that EvoAug-TF achieves comparable results to the original EvoAug package.
• The new package facilitates improved generalization and interpretability in genomic predictions.

Conclusions:

• EvoAug-TF democratizes advanced DNN augmentation techniques for TensorFlow users in genomics.
• The tool addresses the data limitations in functional genomics, improving predictive model performance.
• EvoAug-TF is available as open-source, promoting accessibility and reproducibility in computational biology research.