The surface of nanoparticles changes immediately after intravenous injection because blood proteins adsorb on the surface. How this interface changes during circulation and its impact on nanoparticle distribution within the body is not understood. Here, we developed a workflow to show that the evolution of proteins on nanoparticle surfaces predicts the biological fate of nanoparticles in vivo. This workflow involves extracting nanoparticles at multiple time points from circulation, isolating the proteins off the surface and performing proteomic mass spectrometry. The mass spectrometry protein library served as inputs, while blood clearance and organ accumulation were used as outputs to train a supervised deep neural network that predicts nanoparticle biological fate. In a double-blinded study, we tested the network by predicting nanoparticle spleen and liver accumulation with upward of 94% accuracy. Our neural network discovered that the mechanism of liver and spleen uptake is due to patterns of a multitude of nanoparticle surface adsorbed proteins. There are too many combinations to change these proteins manually using chemical or biological inhibitors to alter clearance. Therefore, we developed a technique that uses the host to act as a bioreactor to prepare nanoparticles with predictable clearance patterns that reduce liver and spleen uptake by 50% and 70%, respectively. These techniques provide opportunities to both predict nanoparticle behavior and also to engineer surface chemistries that are specifically designed by the body.
Keywords: artificial intelligence; machine learning; mass spectrometry; nanoparticles; neural networks; predictive biology; protein corona.