Predicting the protein corona on nanoparticles using random forest models with nanoparticle, protein, and experimental features

Abstract

Nanoparticles (NPs) present in any biological environment form a “corona” of proteins on the NP surface. This protein corona, rather than the bare NP, determines the biological response to the protein–NP complex. Experiments, especially proteomics, can provide an inventory of proteins in the corona, but researchers currently lack a method to predict which proteins will interact with NPs. The ability to predict the protein corona would aid the design of NPs by decreasing the time and cost of experiments. We describe the development and use of random forest regression and classification models to predict protein abundance and enrichment, respectively, on the surface of NPs using a dataset of NP, protein, and experimental features. These models were trained using data generated in-house through the synthesis and functionalization of NPs with varied core material, surface ligand, diameter, and zeta potential. NPs were incubated with fetal bovine serum, a common protein source for cultured cells, to form a corona, which was characterized by proteomics. Both models identified protein abundance in the serum used to form the corona as the most significant predictor of corona proteins. NP zeta potential and hydrodynamic diameter emerged as the most important NP factors. The random forest regression model was used to test the ability to predict the protein corona of NPs that were excluded from the training data. We highlight the best and worst predictions. These findings offer a machine learning approach to guide experiments.