Protein Representation Learning by Geometric Structure Pretraining

Explore protein representation learning through geometric structure pretraining in this 53-minute talk by Zuobai Zhang from Valence Labs. Dive into the world of protein function prediction and discover GearNet, a simple yet effective protein structure encoder. Learn about edge message passing, GearNet-Edge, and various experimental setups. Understand geometric pretraining techniques, including contrastive learning with SimCLR and self-prediction baselines. Examine the results of pretraining on different datasets and gain insights into future research directions. Conclude with a Q&A session to deepen your understanding of this innovative approach to protein representation learning.

- Intro
- Protein Function Prediction
- GearNet: A Simple, Effective Protein Structure Encoder
- Edge Message Passing
- GearNet-Edge
- Experimental Setup
- Geometric Pretraining
- Contrastive Learning: SimCLR
- Self-Prediction Baselines
- Experimental Setup
- Pretraining on Different Datasets
- Conclusion & Future Work
- Q+A