New Frontiers for AI-assisted Design and Discovery in Molecular Generation

Explore the cutting-edge developments in AI-assisted design and discovery of molecules with desirable properties in this 58-minute lecture by Vikas Garg at the Finnish Center for Artificial Intelligence. Delve into the challenges of searching vast combinatorial spaces, modeling molecular interactions, and segregating desirable from undesirable properties. Discover novel approaches to generate high-quality molecules without validity checks, including Modular Flows, Temporal Graph Networks, and symmetry-induced disentanglement on graphs. Learn about the potential applications in drug discovery, materials science, and battery development from an expert in the field of AI and molecular design.

Intro
Motivation
We need to overcome some challenges
We will focus on small molecules
Agenda for today
Background: Generative Models
From Flows to "Modular Flows"
Modular Flows: Overall Picture
Modular Flows: Results
Temporal Graph Networks (TGNs)
PINT: Leveraging the best of both worlds!
SID: Disentangling the factors in graphs
Conclusion
Questions?
Conditional disentanglement improves performance