Advancing Open-Source AI in Chemistry and Materials—From Foundation Models to Integrated Frameworks to Solve Global Challenges

Abstract

This presentation highlights AI advancements in chemistry and material science, emphasizing open-source tools and applications. We will introduce the AI Alliance, a community dedicated to open innovation in AI technology, fostering responsible innovation while ensuring scientific rigor, trust, safety, security, diversity, and economic competitiveness. Particularly, the AI for Chemistry and Materials focuses on developing open-source foundation models for materials. We will highlight the first large structured state space sequence models (SSMs) for molecules, pre-trained on 91 million SMILES samples from PubChem, equating to 4 trillion molecular tokens. This model excels in molecular property prediction, classification, and reconstruction. However, despite advances in computational chemistry and machine learning, many tools remain underutilized due to their complexity and the need for programming skills. We will show how LLM-based AI agents can bridge this gap by orchestrating workflows and multi-step tasks and by integrating a large variety of cheminformatics tools and available foundation models. Finally, we will showcase how these AI technologies can help solve urgent global challenges we are facing, such as the widespread efforts to replace PFAS compounds, or so-called forever chemicals.