(676c) Predicting Protein Adsorption to Engineered Nanoparticles with Supervised Machine Learning Models

Engineered nanoparticles are poised to transform sensing, imaging, and delivery in biological systems.^1,2 In particular, single-walled carbon nanotubes (SWCNTs) are uniquely suited for biological sensing and imaging due to the tissue-transparent and photostable near-infrared fluorescence that they exhibit.¹ As such, SWCNTs have been functionalized with various biomolecules, including synthetic peptides³ and proteins,⁴ to construct optical nanosensors. Optimizing these biomolecule-nanoparticle interactions is key in enhancing nanotechnology function. Yet, functionalized SWCNTs and nanotechnologies more broadly suffer from unpredictable and often unfavorable interactions with the biological environments in which they are applied. When nanoparticles are introduced into biological systems, endogenous proteins rapidly bind and often lead to decreased ability of the nanoparticles to perform their intended functions.^5,6

In this work, we develop a classifier to predict protein-nanoparticle association. Our purpose is two-fold: as one objective, we aim to predict protein-nanoparticle interactions in full biological environments, informing implementation of appropriate anti-biofouling strategies or exploitation of the adsorbed proteins towards effective biological outcomes. Secondly, we intend to predict high-affinity protein binders to nanoparticles to improve the process of protein-nanoparticle construct design. Toward these ends, we build our classifier based on mass spectrometry-based proteomic data that quantitatively characterizes proteins bound to our SWCNT-based nanosensors.⁵ Using widely available protein data, we construct a protein property database and next train and validate a random forest classifier to predict whether proteins adsorb to the nanoparticles. We identify distribution changes among the most important protein properties driving binding. Lastly, we apply the classifier to rapidly identify candidate proteins with high binding affinity to SWCNTs and experimentally validate adsorption of these proteins with a kinetic exchange assay to evaluate the predictive power of our model.⁶ In sum, this classifier serves as a valuable method to both overcome the high failure rate in translating nanotechnologies from in vitro validation to in vivo deployment, and to aid in rational design of future nano-bio tools.

References

1. Hong, G., Antaris, A. L. & Dai, H. Near-infrared fluorophores for biomedical imaging. Nature Biomedical Engineering 1, 1–22 (2017).
2. Mitchell, M. J. et al. Engineering precision nanoparticles for drug delivery. Nature Reviews Drug Discovery 1–24 (2020) doi:10.1038/s41573-020-0090-8.
3. Chio, L. et al. Electrostatic Assemblies of Single-Walled Carbon Nanotubes and Sequence-Tunable Peptoid Polymers Detect a Lectin Protein and Its Target Sugars. Nano Lett. 19, 7563–7572 (2019).
4. Pinals, R. L. et al. Rapid SARS-CoV-2 Spike Protein Detection by Carbon Nanotube-Based Near-Infrared Nanosensors. Nano Lett. (2021) doi:10.1021/acs.nanolett.1c00118.
5. Pinals, R. L. et al. Quantitative Protein Corona Composition and Dynamics on Carbon Nanotubes in Biological Environments. Angewandte Chemie International Edition 59, 23668–23677 (2020).
6. Pinals, R. L., Yang, D., Lui, A., Cao, W. & Landry, M. P. Corona Exchange Dynamics on Carbon Nanotubes by Multiplexed Fluorescence Monitoring. J. Am. Chem. Soc. 142, 1254–1264 (2020).