dZiner

An agentic framework for rational inverse design of materials by replicating and incorporating the expertise of human domain experts across various design tasks and target properties.

The model starts by inputting the material's initial structure as a textual representation. The AI agent dynamically retrieves design guidelines for Property X from scientific literature, the Internet or other resources. Based on these domain-knowledge of guidelines, and any additional design constraints provided in natural language, the agent proposes a new candidate and assesses its chemical feasibility in real-time. Next, it estimates Property X for the new candidate, incorporating epistemic uncertainty, using a cost-efficient surrogate model. Optionally, as part of a human-in-the-loop process, the human chemist can review the agent's new candidates and chain-of-thoughts, providing feedback and suggesting further modifications or constraints, creating an opportunity for human-AI collaboration to guide the exploration process. The agent continues exploring the chemical space, guided by chemistry-informed rules, until it meets the convergence criteria.

Human-in-the-loop Inverse Design

Collaborative efforts between a human expert and AI agents hold significant promise. In the case of molecular design for WDR5 ligands, we examined human guidance to refine the modifications based on docking scores and structural generation.

user-attachments/assets

Closed-loop Inverse Design

We applied dZiner to the rational inverse design of likely synthesizable organic linkers for metal-organic frameworks with high CO2 adsorption capacity at 0.5 bar of pressure. These MOFs come with pcu topology and three types of inorganic nodes: Cu paddlewheel, Zn paddlewheel, and Zn tetramer (three most frequent node-topology pairs in the hMOF dataset). Design constraints such as keeping molecular weight lower than 600 g/mol and excluding certain potentially unstable functional groups (nitrosylated, chloro-, fluoro- amines) are simply added to the model as natural language text.

Benchmarks

How Can I Use dZiner for My Own Materials Inverse Design Problem?

dZiner can work with different textual representation for materials. You can even apply your own surrogate model to your own materials inverse design problem. Here are some example notebooks that can help you get started:

• Quick Start

• Example 1 - MOF Organic Linker Design for CO2 Adsorption

• Example 2 - Desigining Peptides with a Lower Hemolytic Activity

• Example 3 - Low CMC Surfactant Design

• Example 4 - Different Choices of Surrorgate Models

• Example 5 - WDR5 Ligand Inverse Design

• Example 6 - Human-in-the-loop design of more soluble molecules with IBM’s embedding model (smi-TED)

Citation

For more details checkout preprint and if you are using our methodology, please consider citing us using the citation below:

@misc{ansari2024dzinerrationalinversedesign,
      title={dZiner: Rational Inverse Design of Materials with AI Agents}, 
      author={Mehrad Ansari and Jeffrey Watchorn and Carla E. Brown and Joseph S. Brown},
      year={2024},
      eprint={2410.03963},
      archivePrefix={arXiv},
      primaryClass={physics.chem-ph},
      url={https://arxiv.org/abs/2410.03963}, 
}