Explore a 44-minute conference talk on Conditional Graph Information Bottleneck (CGIB) for molecular relational learning. Dive into the innovative approach of detecting core subgraphs to predict interactions between molecular pairs. Learn how CGIB mimics chemical reactions by finding minimal sufficient information in substructures conditioned on paired molecules. Discover the background, methodology, and experimental results of this novel framework that outperforms state-of-the-art baselines in various molecular science applications. Gain insights from the speaker, Namkyeong Lee of Valence Labs, and engage with the AI drug discovery community through the provided Portal link. Follow the talk's structure from introduction to Q&A session, understanding how CGIB addresses the overlooked nature of chemistry in existing molecular relational learning methods.
Overview
Syllabus
- Intro
- Background
- Methodology
- Experiments
- Conclusion
- Q+A
Taught by
Valence Labs
Related Courses
-
Equivariant 3D-Conditional Diffusion Models for Molecular Linker Design
-
Removing Biases from Molecular Representations via Information Maximization
-
Large Language Molecular Representation and Learning
-
Interpretable Chirality-Aware Graph Neural Networks for QSAR Modeling in Drug Discovery
-
Structure-Aware Protein Self-Supervised Learning
