AI-Based High-Throughput Screening of Thiol-Ene/Yne Reactions for Self-Healing Materials Applications (ASTRA)

Modern society's dependence on plastics and polymeric materials generates approximately 380 million tonnes of plastic waste annually, with significant environmental and economic consequences. Unlike biological materials that possess intrinsic self-repair capabilities, conventional polymers suffer from irreversible damage that limits their lifespan and necessitates premature replacement. Self-healing materials offer a solution by enabling autonomous or stimuli-triggered repair of damage, potentially extending material lifetimes by years or decades and dramatically reducing waste.

Self-healing polymers based on reversible covalent bonds, such as those in thiol-ene and thiol-yne reactions, represent one of the most promising approaches, as they combine mechanical robustness with repairability. However, to this date the discovery of optimal thiol-ene/yne pairs remains largely empirical, relying on time-consuming and resource-intensive trial-and-error approaches. Here, computational approaches promise new opportunities for rational, quantitative design of such reaction pairs.

Quantum chemical methods, particularly Density Functional Theory (DFT), can in principle predict reaction barriers and thermodynamics with useful accuracy. However, such calculations remain prohibitively expensive for large-scale screening of reaction spaces containing tens of thousands of combinations. Recent advances in machine learning (ML) for chemistry open new possibilities, but also require extensive, high-quality training data.

Here, multi-tier ML architectures based on training data from different levels of theory promise new opportunities to identify optimal reaction pairs for thiol-ene/yne reactions. With this, we aim for data-driven design of new self-healing polymers for specific materials applications.