Structure-Based Multilevel Descriptors for High-throughput Screening of Elastomers.

Area of Science:

• Materials Science
• Polymer Chemistry
• Computational Materials Science

Background:

• High-throughput screening (HTS) for new materials, particularly elastomers, relies on accurate property prediction using machine learning (ML).
• Existing descriptors often lack universally available data for all potential elastomer candidates, limiting HTS efficiency.
• A need exists for universally applicable descriptors derived directly from molecular structure.

Purpose of the Study:

• To introduce novel structure-based multilevel (SM) descriptors for elastomers.
• To demonstrate the capability of SM descriptors in accurately predicting elastomer properties.
• To establish HTS pipelines for elastomer discovery using these new descriptors.

Main Methods:

• Development of hierarchical SM descriptors capturing local and global elastomer structures from molecular data.
• Utilized the SM-Morgan Fingerprint (SM-MF) descriptor within an ML model.
• Established HTS pipelines for screening elastomers based on targeted toughness, critical strain, and Young's modulus.

Main Results:

• The SM-MF descriptor achieved a high accuracy of 0.91 in predicting elastomer toughness.
• Successful implementation of HTS pipelines for screening elastomers with specific mechanical properties.
• Demonstrated the broad applicability of SM descriptors across different material property targets.

Conclusions:

• SM descriptors offer a universally applicable and computationally efficient method for elastomer characterization.
• These descriptors significantly enhance the potential of HTS for discovering novel elastomers.
• The user-friendly nature of SM descriptors facilitates broader adoption in materials discovery.