Decoding Immunomodulatory Hydrogels for Arthritis: Comparative Insights from Predictive Machine Learning and Large Language Models

Abstract

Hydrogels are increasingly recognized as promising therapeutics for arthritic joints, extending their traditional role as mechanical lubricants to modulators of joint immunity. However, the rational design of these materials remains challenging, with progress largely driven by empirical experimentation. To address this, we curated a comprehensive database of 220 hydrogel formulations from 317 published studies and applied an interpretable machine learning (ML) framework to uncover the relationships between hydrogel design parameters and the arthritis severity score. Using a Random Forest algorithm, our model achieved an external validation accuracy of 0.67 in predicting effective hydrogel therapies for arthritis. Analysis revealed a clear hierarchy of design principles: the choice of functional agent, base polymer, and elastic modulus were the most influential predictors of therapeutic efficacy, with composite agents, protein-based polymers, and softer hydrogels most strongly associated with positive therapeutic outcomes. Mechanistic investigations further demonstrated that successful hydrogels promote an anti-inflammatory M2 macrophage phenotype. Benchmarking against classical statistical methods and a large language model framework showed that our ML approach provided more robust, nuanced insights into complex feature interactions. This data-driven framework offers a generalizable blueprint for the rational design of next-generation immunomodulatory hydrogels, paving the way for more effective arthritis therapies.