POLYGEN · New Paradigms for Deep Generative Modeling of Polymers

Polymers are the most ubiquitous materials class in modern society. Most of us spend a significant part of our lives in buildings made of polymers, using electronics manufactured with special polymers, and taking medicines encapsulated in carefully-optimized polymer formulations. In many cases, the design of new polymers is key to addressing global challenges in sustainability, health, and security. However, current polymer design methods rely heavily on trial-and-error experimentation and are grounded on imprecise assumptions about their structure, which is not only time-consuming and costly, but also offers limited insights into the relationship between structure, synthesis, and properties beyond the simplest polymers. While data-driven methods for small molecule engineering have flourished in recent years, artificial intelligence (AI) methods designed for polymers remain severely underdeveloped due to the unique challenges they pose. Their design involves navigating a noisy data space in search of compounds satisfying a complex set of specifications, made even more challenging when we’re interested in sampling novel, unique materials. The lack of well-defined, regular structures in polymers, coupled with a vastly different synthesis space, has also led to inefficient representations for them and an overreliance on hand-crafted simplifications, necessitating the development of information-rich yet scalable representa-tions for diverse polymer classes. I will tackle these gaps in AI-guided polymer design through the de-velopment of new paradigms for (1) learnable polymer representations and (2) generative AI frameworks for polymer design and synthesis. The methods I will develop bridge unique ideas from polymer science and AI to create a ground-breaking new approach for engineering polymers to specification and transform the current polymer design landscape across many industries via scalable data-driven strategies.