Machine Learning Revolutionizes Nanoparticle Design for Drug Delivery

Neurodegenerative diseases, which impact millions globally, often face treatment challenges due to the blood-brain barrier (BBB). A research team from the University of the Basque Country in Spain and Tulane University in the USA, led by the CHEMIF.PTML Lab, has developed a groundbreaking machine learning technique aimed at optimizing nanoparticle design for efficient drug delivery to the brain.

The team’s work involves creating nanoparticles capable of crossing the BBB, thereby enhancing drug transport to specific brain regions. Designing these targeted nanoparticles has traditionally been complex and time-consuming, necessitating the development of advanced frameworks to identify candidates with desirable bioactivity profiles.

To tackle this challenge, the researchers employed machine learning, utilizing molecular and clinical data to uncover trends that could lead to innovative drug delivery methods with improved efficiency and minimal side effects. Unlike the slow and costly trial-and-error approaches previously used, machine learning offers a faster screening process for a wide range of nanoparticle compositions. However, traditional methods often struggle due to the lack of suitable datasets.

To overcome this limitation, the CHEMIF.PTML Lab introduced the IFE.PTML method, which integrates information fusion, Python-based encoding, and perturbation theory with machine learning algorithms. According to corresponding author Humberto González-Díaz, “The main advantage of our IFE.PTML method lies in its ability to handle heterogeneous nanoparticle data.” This approach merges diverse data types—such as physicochemical properties and bioassays—while applying perturbation theory to model uncertainties, resulting in more reliable predictions of nanoparticle behavior.

The research team constructed a comprehensive database containing physicochemical and bioactivity parameters for 45 different nanoparticle systems across 41 cell lines. Using this data, they trained the IFE.PTML models employing three machine learning algorithms: random forest, extreme gradient boosting, and decision tree. The random forest-based model demonstrated the highest performance, achieving accuracies of 95.1% on training datasets and 89.7% on testing datasets.

Experimental Validation of Predictions

To validate the applicability of their model, the researchers synthesized two novel magnetite nanoparticle systems, measuring 31 nm in diameter (Fe3O4_A) and 26 nm (Fe3O4_B). These magnetite-based nanoparticles are biocompatible, easily functionalized, and possess a high surface area-to-volume ratio, making them efficient drug carriers. The team coated the nanoparticles with either PMAO (poly(maleic anhydride-alt-1-octadecene)) or PMAO combined with PEI (poly(ethyleneimine)) to enhance water solubility.

The team characterized the structural, morphological, and magnetic properties of the four nanoparticle systems. They subsequently used the optimized model to predict the likelihood of favorable bioactivity for drug delivery in various human brain cell lines, including models of neurodegenerative diseases, brain tumors, and the BBB.

The inputs for their model included a reference function based on bioactivity parameters and perturbation theory operators for several nanoparticle parameters. The IFE.PTML model calculated critical bioactivity parameters, concentrating on indicators of toxicity, efficacy, and safety. These included the 50% cytotoxic, inhibitory, lethal, and toxic concentrations, along with the zeta potential, which influences the nanoparticles’ ability to cross the BBB. The model output binary results: “0” for undesired and “1” for desired bioactivities.

The analysis revealed that PMAO-coated nanoparticles emerged as the most promising candidates for applications involving the BBB and neuronal interactions, due to their favorable stability and biocompatibility. In contrast, nanoparticles with PMAO-PEI coatings showed potential for targeting brain tumor cells.

The researchers noted that, where comparisons were feasible, the trends predicted by the random forest-based IFE.PTML model aligned with experimental findings and previous studies documented in the literature. This congruence suggests that the model is both efficient and robust, providing valuable insights into nanoparticle-coating combinations aimed at specific targets.

González-Díaz stated, “The present study focused on the nanoparticles as potential drug carriers. Therefore, we are currently implementing a combined machine learning and deep learning methodology with potential drug candidates for neurodegenerative diseases.”

The CHEMIF.PTML Lab operates as a multicentre laboratory of the Biofisika Institute of UPV/EHU and the Spanish National Research Council, with support from Ikerbasque, the Basque Foundation for Science.