Researchers have utilized machine learning techniques alongside supercomputer simulations to explore the interactions between tiny gold nanoparticles and blood proteins. Their findings revealed that machine learning models trained on detailed atom-scale molecular dynamics simulations can effectively predict beneficial interactions between nanoparticles and proteins. This innovative approach paves the way for simulating the effectiveness of gold nanoparticles as targeted delivery systems in precision nanomedicine.
Researchers in the Nanoscience Center at the University of Jyväskylä, Finland, have utilized machine learning techniques alongside supercomputer simulations to explore the interactions between tiny gold nanoparticles and blood proteins. Their findings revealed that machine learning models trained on detailed atom-scale molecular dynamics simulations can effectively predict beneficial interactions between nanoparticles and proteins. This innovative approach paves the way for simulating the effectiveness of gold nanoparticles as targeted delivery systems in precision nanomedicine.
Hybrid nanostructures that combine biomolecules with inorganic nanomaterials represent a largely untapped area of research, offering potential for unique applications in bioimaging, biosensing, and nanomedicine. For the development of these applications, a deep understanding of the dynamic properties at the nano-bio interface is essential. However, accurately modeling the attributes of this interface is challenging, as crucial processes—like electronic charge transfer, chemical reactions, or surface restructuring of biomolecules—can occur across varying lengths and timeframes. Additionally, these atomistic simulations must be performed in a suitable aqueous environment.
Utilizing machine learning to investigate atomic-level interactions
Recently, researchers from the University of Jyväskylä demonstrated a significant acceleration in atomistic simulations for studying interactions between metal nanoparticles and blood proteins. Using extensive data from molecular dynamics simulations involving gold nanoparticles and proteins in water, they implemented graph theory and neural networks to develop a method that predicts the most suitable binding sites of nanoparticles to five common human blood proteins (serum albumin, apolipoprotein E, immunoglobulin E, immunoglobulin G, and fibrinogen). The predictions made by the machine learning models were successfully confirmed through long-timescale atomistic simulations.
– In recent months, we published a computational study indicating the potential to selectively target over-expressed proteins on cancer cell surfaces using functionalized gold nanoparticles loaded with peptides and cancer drugs, explains Hannu Häkkinen, professor of computational nanoscience. With our new machine learning methodology, we can extend our research to examine how drug-laden nanoparticles engage with blood proteins and how these interactions might influence the effectiveness of the drug carriers, Häkkinen adds.
Future Research Directions
The findings will facilitate further investigation aimed at developing new computational methods for studying interactions between metal nanoparticles and biomolecules.
“Machine learning will be a vital asset when assessing the role of nanoparticles in diagnostics and therapeutic applications within nanomedicine. This will be a primary focus of our upcoming project ‘Dynamic Nanocluster–Biomolecule Interfaces,’ which is backed by the European Research Council,” Häkkinen remarked.
The findings were published in two scholarly articles in the international journals, Advanced Materials and Bioconjugate Chemistry. This research was supported by the EuroHPC funding program from the Research Council of Finland. The computational resources utilized were provided by the Finnish Grand Challenge Projects BIOINT and NanoGaC, leveraging the LUMI and Mahti supercomputers, respectively, hosted at the Finnish supercomputing center CSC.
