Meta-Analysis and Machine Learning Prediction of Protein Corona Composition across Nanoparticle Systems in Biological Media
Alexa Canchola, Keyuan Li, Kunpeng Chen, Alejandro Borboa-Pimentel, C.J. Chou, R. Rama, Chi‐Yun Chen, Xinyue Chen, Michael Strobel, Jim E. Riviere, Nancy A. Monteiro‐Riviere, Mingxun Wang, Fan Zhang, Zhoumeng Lin, Wei‐Chun Chou
Abstract
A comprehensive understanding of protein corona (PC) composition is critical for engineering nanoparticles (NPs) with optimal safety and therapeutic performance, because the PC governs NP pharmacokinetics, biodistribution, and cellular interactions. Yet systematic analyses are hampered by the absence of standardized, richly annotated data sets. Here, we introduce the Protein Corona Database (PC-DB), which compiles data from 83 studies (2000-2024) and integrates 817 NP formulations with quantitative profiles of 2497 adsorbed proteins. The PC-DB exposes pronounced heterogeneity in NP materials (metal 28.8%, silica 22.8%, lipid-based 14.8%), surface modifications, sizes (1-1400 nm), and ζ-potentials (-70 to +70 mV). Subsequent meta-analysis shows that silica, polystyrene, and lipid-based NPs smaller than 100 nm with moderately negative to neutral ζ-potentials preferentially bind the lipoproteins APOE and APOB-100, which are linked to receptor-mediated uptake and enhanced delivery efficiency. In contrast, metal and metal-oxide NPs carrying highly negative surface charge enrich complement component C3, indicating a greater likelihood of immune recognition and clearance. Interpretable machine learning models (LightGBM and XGBoost; ROC-AUC > 0.85) confirm NP size, ζ-potential, and incubation time as the most influential predictors of protein adsorption. These results delineate how physicochemical parameters dictate PC composition and illustrate the power of predictive modeling to guide rational NP design.