Mentors: Zilu Zhang, Dr. Daniel Reker PhD
Department of Biomedical Engineering, Duke University
Co-aggregating nanoparticles can stabilize drugs with more than 90% drug loading capacity. While machine learning can be productively employed to identify nanoparticles, this approach requires large datasets. Simulations provide an opportunity to design nanoparticles without prior data generation, but this method has not yet shown sufficient accuracy. Here, I will develop a novel simulation-based approach that achieves productive accuracy of nanoparticle predictions. By pairing a predictive machine learning model and molecular dynamics simulation software, we analyzed hydrogen bond formation in simulations and used our findings to identify pairs of interest. We compared our predictions against already known data and found that the presence of hydrogen bonding indicates higher likeliness of nanoparticle formation in more than 75% of analyzed pairs. Using this analysis protocol, we plan to analyze and predict other small antiviral nanoparticle formulations aimed at targeting viral diseases such as COVID-19.