Reproduce “Enhancing Protein Predictive Models via Proteins Data Augmentation: A Benchmark and New Directions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rui Sun, Lirong Wu, Haitao Lin, Yufei Huang, Stan Z. Li
This paper presents a comprehensive study of data augmentation techniques for protein sequences and introduces an Automated Protein Augmentation (APA) framework. The framework, which includes novel semantic-level augmentation methods, consistently improves protein predictive model performance across various tasks and architectures, showing an average 10.55% increase over models without augmentation.
Key insights
- Protein data augmentation can significantly improve predictive model performance, even for tasks with limited labeled data.
- Semantic-level augmentation techniques, like Integrated Gradients Substitution and Back Translation, that incorporate biological properties generally outperform generic data augmentation methods.
- Data augmentation and pre-training are complementary strategies, with augmentation providing further performance gains for pre-trained protein language models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.