Can LLMs Predict Polymer Physics Just by Reading Synthesis and Processing Prose?

2026-05-07

AI Summary

This work addresses the limitations of conventional polymer property prediction models, which rely solely on chemical structures (e.g., SMILES) and neglect critical experimental context such as synthesis pathways, processing histories, and testing conditions, thereby failing to accurately capture real-world performance. To overcome this, the authors propose PolyLM—a novel framework that predicts physical, mechanical, and thermal properties of polymers directly from full-text scientific literature using only natural language, without any structured chemical input. Built upon the 9-billion-parameter Qwen3.5-9B large language model and fine-tuned with LoRA and task-level uncertainty weighting, PolyLM is trained on a large-scale dataset of 185,000 papers. Evaluated on 68,283 held-out samples, it achieves a median R² of 0.74 across 22 properties, with several key metrics exceeding R² = 0.80, substantially advancing the state of the art in polymer property prediction.

Abstract

Can large language models predict physical and mechanical polymer properties simply by reading unstructured scientific prose? Polymer performance is rarely determined by chemical structure alone; identical nominal polymers can exhibit drastically different behaviors depending on their synthesis route, processing history, morphology, and testing conditions. Yet, state-of-the-art polymer property models typically rely on structure-only representations -- such as SMILES or molecular graphs -- which strip away this vital experimental context. In this work, we introduce \textbf{PolyLM}, a natural-language-only, process- and condition-aware framework that predicts materials performance directly from full-text literature. By circumventing structural inputs entirely, PolyLM preserves the nuanced, unstructured descriptions of synthesis and processing reported by domain scientists. To train this framework, we curated an unprecedented, literature-scale dataset encompassing 185,000 scientific papers and over 276,400 unique polymer samples across 22 physical, mechanical, and thermal properties. We fine-tuned a massive 9-billion-parameter language model (Qwen3.5-9B) using Low-Rank Adaptation (LoRA) and task-level uncertainty weighting. Evaluated on 68,283 held-out observations, the model achieves remarkably high predictive accuracy, establishing new state-of-the-art benchmarks for complex properties. Across the 22 diverse targets, the model achieves a median $R^2$ of 0.74, with predictions for key thermal, mechanical, and physicochemical properties frequently surpassing an $R^2$ of 0.80. These results unequivocally demonstrate that natural language is a powerful, highly scalable interface for realistic materials performance prediction.

Problem

• polymer property prediction
• natural language processing
• synthesis and processing
• materials informatics
• unstructured scientific text

Innovation

• natural language processing
• polymer property prediction
• process-aware modeling
• large language models
• materials informatics