Development of Foundational AI Agents

THRUST 1

In MMLI 2.0, the overall goal of Thrust 1 is to develop foundational AI agents for discovery and synthesis of functional molecules, with a focus on three key areas:

A function- and synthesis-aware modular chemical language model (mCLM) to enable the prediction of new molecules of emergent functions derived from their modules, and all of the AI-generated molecules will be accessible via automated assembly by non-specialists.
Knowledge-augmented theme-specific LLMs (themeLLM) for hypothesis generation and experimental design.
Innovative AI Agents with critical thinking to enable the development of other frontier AI tools such as generative AI within Thrusts 2-4.

CURRENT PROJECTS

mCLM: Function- and Synthesis-aware Modular Chemical Language Models

Incorporation of both function- and synthesis-related knowledge into the small molecule tokenization process a priori.

Development of Theme-Specific LLM Agents for Chemistry

Development of ThemeLLM, a theme-specific knowledge-augmented large language model focusing on specific research topic

Innovative AI Agents with Critical Thinking

Development of agent systems drawing on specialized models and tools to execute complex tasks in a semi-automatic way while keeping humans in the loop for feedback and higher-level guidance.

Leveraging NCSA Supercomputing Resources to Develop a MMLI Database

Transition cyberinfrastructure to production, providing an expanding user base with real-time, self-serve access to datasets and compute resources