Data-driven discovery of design principles of immunomodulatory hydrogels in arthritis treatment

Authors:

Abstract
Hydrogels are evolving from simple mechanical lubricants to active modulators of joint immunity in arthritis therapy. However, biomaterial development remains heavily constrained by empirical trial-and-error and a tendency to follow prevailing research trends. To challenge this dogma, we established a data-driven framework that bridges retrospective literature mining with prospective experimental validation. An interpretable machine learning (ML) model, trained on a comprehensive database of 220 hydrogel formulations from 317 published studies, was used to decode key design principles. The model identified the base polymer, functional agent, and elastic modulus as the most critical predictors of therapeutic efficacy, with immune profiling revealing that this therapeutic benefit is predominantly driven by anti-inflammatory M2 macrophage polarization. Crucially, while retrieval-augmented large language models inherited literature popularity bias by favoring trending polysaccharides, our ML framework and subsequent prospective assays prioritized protein-based matrices. Prospective experimental validation demonstrated that material composition outweighs mechanical properties, and that protein-based hydrogels induce stronger therapeutic M2 polarization than the currently trending polysaccharide-based alternatives. By challenging existing design assumptions, this work provides a generalizable strategy to transition biomaterials engineering from trend-following empirical testing to rational, data-driven discovery.