Inverse-designing Polymers with Unified Graph-grammar Learning
Abstract
The traditional method of identifying polymers with the desired physical properties is a trial-and-error process that is time-consuming and inefficient. While machine learning methods have made progress in accelerating the process of discovering materials, the current state-of-the-art algorithms are still data-intensive and require large amounts of training data. This can be particularly challengingfor complex polymers, which often have a large molecular design space and-or require resource-intensive simulations or experiments to study their structure-property relationships. As a result, obtaining a representative and sufficiently large dataset is a bottleneck for designing such polymers.To address this issue, we propose a data-efficient framework that uses graph grammar and reinforcement learning to design polymer molecules with targeted physical properties. This framework interprets molecules as graph networks, with atoms represented as nodes and bonds represented as edges. By predicting a set of rules (i.e., grammar) for constructing and modifying these molecular graphs in a bottom-up fashion, we can tailor the physical properties of the molecules to meet specific requirements. These rules could specify how to add or remove atoms, change the types of atoms or bonds, or modify the arrangement of atoms in the molecule. The overarching emphasis of this framework is on achieving extreme data efficiency while maintaining accuracy in achieving targeted physicalproperties. The successful outcome of the project will bypass the slow process of identifying suitable polymer candidates and directly kick-start in-depth computational and experimental exploration of properties like mechanical behavior, manufacturability, recyclability, and more.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 22, 2024
- Source ID
- FA86552317020
Entities
People
- Siddhant Kumar
Organizations
- Air Force Office of Scientific Research
- Delft University of Technology
- United States Air Force