Intelligent Materials Modelling: Large Language Models Versus Partial Least Squares Regression for Predicting Polysulfone Membrane Mechanical Performance
arXiv:2603.13834v1 Announce Type: new Abstract: Predicting the mechanical properties of polysulfone (PSF) membranes from structural descriptors remains challenging due to extreme data scarcity typical of experimental studies. To investigate this issue, this study benchmarked knowledge-driven inference using four large language models (LLMs) (DeepSeek-V3, DeepSeek-R1, ChatGPT-4o, and GPT-5) against partial least squares (PLS) regression for predicting Young's modulus (E), tensile strength (TS), and elongation at break (EL) based on pore diameter (PD), contact angle (CA), thickness (T), and porosity (P) measurements. These knowledge-driven approaches demonstrated property-specific advantages over the chemometric baseline. For EL, LLMs achieved statistically significant improvements, with DeepSeek-R1 and GPT-5 delivering 40.5% and 40.3% of Root Mean Square Error reductions, respectively, reducing mean absolute errors from $11.63\pm5.34$% to $5.18\pm0.17$%. Run-to-run variability was mark
arXiv:2603.13834v1 Announce Type: new Abstract: Predicting the mechanical properties of polysulfone (PSF) membranes from structural descriptors remains challenging due to extreme data scarcity typical of experimental studies. To investigate this issue, this study benchmarked knowledge-driven inference using four large language models (LLMs) (DeepSeek-V3, DeepSeek-R1, ChatGPT-4o, and GPT-5) against partial least squares (PLS) regression for predicting Young's modulus (E), tensile strength (TS), and elongation at break (EL) based on pore diameter (PD), contact angle (CA), thickness (T), and porosity (P) measurements. These knowledge-driven approaches demonstrated property-specific advantages over the chemometric baseline. For EL, LLMs achieved statistically significant improvements, with DeepSeek-R1 and GPT-5 delivering 40.5% and 40.3% of Root Mean Square Error reductions, respectively, reducing mean absolute errors from $11.63\pm5.34$% to $5.18\pm0.17$%. Run-to-run variability was markedly compressed for LLMs ($\leq$3%) compared to PLS (up to 47%). E and TS predictions showed statistical parity between approaches ($q\geq0.05$), indicating sufficient performance of linear methods for properties with strong structure-property correlations. Error topology analysis revealed systematic regression-to-the-mean behavior dominated by data-regime effects rather than model-family limitations. These findings establish that LLMs excel for non-linear, constraint-sensitive properties under bootstrap instability, while PLS remains competitive for linear relationships requiring interpretable latent-variable decompositions. The demonstrated complementarity suggests hybrid architectures leveraging LLM-encoded knowledge within interpretable frameworks may optimise small-data materials discovery.
Executive Summary
This study compared the predictive capabilities of large language models (LLMs) and partial least squares (PLS) regression for predicting the mechanical properties of polysulfone (PSF) membranes. The results demonstrate that LLMs excel in predicting non-linear properties, such as elongation at break, while PLS regression remains competitive for linear properties like Young's modulus and tensile strength. The study highlights the complementarity of LLMs and PLS regression, suggesting that hybrid architectures leveraging LLM-encoded knowledge within interpretable frameworks may optimize small-data materials discovery. The findings have significant implications for the materials science community, particularly in addressing the challenges associated with data scarcity in experimental studies. Furthermore, the study's results have the potential to inform the development of more efficient and accurate predictive models for various materials applications.
Key Points
- ▸ LLMs outperform PLS regression in predicting non-linear properties like elongation at break
- ▸ PLS regression remains competitive for linear properties like Young's modulus and tensile strength
- ▸ Hybrid architectures leveraging LLM-encoded knowledge may optimize small-data materials discovery
Merits
Advancements in Materials Science
The study contributes to the development of more accurate predictive models for materials properties, which can inform the design and optimization of materials with specific properties. This has significant implications for various applications, including energy storage, filtration, and biomedical devices.
Methodological Contributions
The study demonstrates the potential of LLMs in materials science and provides a framework for evaluating their performance in comparison to traditional chemometric methods like PLS regression. This has the potential to inform the development of more efficient and effective predictive models for various materials applications.
Demerits
Limited Generalizability
The study's findings may not generalize to other materials systems or properties due to the specificity of the PSF membrane data used in the study. Further research is needed to validate the results and extend the findings to other materials applications.
Computational Resource Intensity
The use of LLMs can be computationally intensive, which may limit their adoption in certain materials science applications. Further research is needed to develop more efficient LLM architectures or to explore alternative methods that can achieve similar performance without the associated computational overhead.
Expert Commentary
The study's findings are significant because they demonstrate the potential of LLMs in materials science, particularly in predicting non-linear properties like elongation at break. However, the results also highlight the limitations of LLMs, including their limited generalizability and computational resource intensity. Further research is needed to validate the results and extend the findings to other materials applications. The study's methodology provides a framework for evaluating the performance of LLMs in comparison to traditional chemometric methods like PLS regression, which has the potential to inform the development of more efficient and effective predictive models for various materials applications.
Recommendations
- ✓ Further research is needed to validate the results and extend the findings to other materials applications.
- ✓ The development of more efficient LLM architectures or alternative methods that can achieve similar performance without the associated computational overhead is recommended.