Exploring high thermal conductivity polymers via interpretable machine learning with physical descriptors

Published in npj Computational Materials: https://www.nature.com/articles/s41524-023-01154-w

The efficient and economical exploitation of polymers with high thermal conductivity (TC) is essential to solve the issue of heat dissipation in organic devices. Currently, the experimental preparation of functional polymers with high TC remains a trial-and-error process due to the multi-degrees of freedom during the synthesis and characterization process. Polymer informatics equips machine learning (ML) as a powerful engine for the efficient design of polymers with desired properties. However, available polymer TC databases are rare, and establishing appropriate polymer representation is still challenging. In this work, we propose a high-throughput screening framework for polymer chains with high TC via interpretable ML and physical feature engineering. The hierarchical down-selection process stepwise optimizes the 320 initial physical descriptors to the final 20 dimensions and then assists the ML models to achieve a prediction accuracy R2 over 0.80, which is superior to traditional graph descriptors. Further, we analyze the contribution of the individual descriptors to TC and derive the explicit equation for TC prediction using symbolic regression. The high TC polymer structures are mostly π-conjugated, whose overlapping p-orbitals enable easy maintenance of strong chain stiffness and large group velocities. Ultimately, we establish the connections between the individual chains and the amorphous state of polymers. Polymer chains with high TC have strong intra-chain interactions, and their corresponding amorphous systems are favorable for obtaining a large radius of gyration and causing enhanced thermal transport. The proposed data-driven framework should facilitate the theoretical and experimental design of polymers with desirable properties.