De Novo Design of Polyimides Leveraging Deep Reinforcement Learning Agent

Area of Science:

• Materials Science
• Organic Chemistry
• Computational Chemistry

Background:

• Designing organic films with superior properties is crucial but challenging due to vast chemical spaces.
• Screening optimal materials for specific applications requires efficient design strategies.

Purpose of the Study:

• To propose a novel deep reinforcement learning-based strategy for de novo, template-free polyimide creation.
• To develop an agent, DAPiGen, capable of designing polyimides with tailored performance characteristics.

Main Methods:

• A fragment-based generation architecture using active fragments from polyimides.
• Integration of property predictors from four machine learning models.
• Deep reinforcement learning to guide the molecular design process.

Main Results:

• Successful de novo creation of polyimides with targeted properties for flexible displays.
• Achieved higher transparency, lower coefficient of linear thermal expansion, superior tensile strength, and elevated glass transition temperature.
• Experimental validation confirmed the efficacy and reliability of the designed polyimides.

Conclusions:

• The DAPiGen strategy provides a scalable approach for the inverse design of polymeric materials.
• This research offers a paradigm for accelerating the discovery of advanced functional materials.
• The methodology can guide future structural engineering endeavors in materials science.