Learning to Group Auxiliary Datasets for Molecule Prediction

Explore a comprehensive lecture on leveraging auxiliary datasets for molecular machine learning. Delve into the challenges of limited annotations in small molecule datasets and learn strategies to address them through collaboration with auxiliary datasets. Understand the concept of negative transfer and its impact on model performance. Discover MolGroup, an innovative approach that separates dataset affinity into task and structure components to predict the potential benefits of auxiliary molecule datasets. Examine the routing mechanism optimized through bi-level optimization and its ability to maximize target dataset performance. Gain insights into empirical analysis, benchmarking results, and the optimal combination of auxiliary datasets for target datasets. Conclude with a Q&A session to further clarify concepts and applications in AI-driven drug discovery.

- Intro + Background
- Auxiliary Molecule Datasets
- Understanding Relationships Between Datasets
- MolGroup: Routing Mechanism
- MolGroup: Bi-Level Optimization
- Benchmarking
- Conclusions
- Q&A