AI-Driven Discovery of Novel Polymers: A Comprehensive Review

Abstract

The discovery and optimization of polymeric materials have traditionally relied on experiment-driven, trial-and-error workflows that are time-consuming and resource intensive. Artificial intelligence (AI) — particularly machine learning (ML), graph neural networks (GNNs), generative models, and active learning — is transforming polymer science by enabling rapid property prediction, inverse design, and closed-loop experimental optimization. This review surveys recent advances (2018–2025) in data resources, polymer representations, predictive and generative AI models, optimization strategies (including Bayesian optimization and active learning), and automated/self-driving laboratories. We highlight landmark platforms (Polymer Genome, Open Macromolecular Genome), methodological progress (multitask GNNs, chemical language models such as polyBERT, and benchmarks for deep generative models), and successful demonstrations of AI-assisted polymer discovery in energy, electronics, healthcare, and sustainable plastics.

Key challenges remain: scarcity and heterogeneity of polymer data, difficulty representing polymer ensembles and architectures, synthetic feasibility of generated candidates, model interpretability, and integration with experimental workflows. We discuss strategies to address these issues — standardized databases and representations, self-supervised learning, synthesis-aware generative models, and tighter AI–robotics integration — and outline opportunities where AI can accelerate green polymer chemistry, circular-economy polymers, and application-directed multi-property optimization. The review concludes by advocating for community efforts in data curation, open benchmarks, and interdisciplinary training to realize AI’s promise for fast, cost-effective polymer innovation.