University of Michigan researchers propose a new machine learning (ML) model that predicts interactions between nanoparticles and proteins

This Article Is Based On The Research 'Unifying structural descriptors for biological and bioinspired nanoscale complexes'. All Credit For This Research Goes To The Researchers of This Project 👏👏👏

Please Don't Forget To Join Our ML Subreddit

A new Machine Learning ML model created at the University of Michigan anticipates interactions between nanoparticles and proteins. More than simple drug delivery vehicles, nanoparticles have been re-evaluated. As a medicine in itself, they are effective.


Nanoparticles are known to be adjuvants, enzyme mimics, and inhibitors of amyloid fibrillation, among other functions. To further improve them, it is necessary to understand their interactions with proteins. Protein-nanoparticle assemblies can be designed using knowledge of protein-protein interactions. However, the chemical and biological inputs used in computational tools for protein-protein interactions do not apply to inorganic nanoparticles.

Geometric and graph-theoretical descriptors are universally applicable to natural and inorganic nanostructures. They can accurately predict interaction sites in protein pairs with a 90% classification probability by analyzing chemical, geometric, and theoretical descriptors of graphs for protein complexes. Machine learning algorithms trained on protein-protein interactions have been extended to inorganic nanoparticles, and a nearly identical correspondence between experimental and anticipated protein interaction sites has been observed. Based on these findings, lock and key complexes can be predicted for different organic and inorganic nanoparticles applied to other chemical structures.

Lock and key complexes have been used to conceptualize protein interactions, depicted in many successful protein-protein interaction (PPI) algorithms. These and other computational tools use the pairwise similarity of a putative “key” to many other “keys” to predict protein complex formation and interaction sites. Nanoparticle (NP)-protein interactions can benefit from a similar approach. However, an X-ray diffraction data library for NP-protein couples comparable to the Protein Data Bank (PDB) is currently lacking.

PPI packages also assume that the interacting molecules are linear polymers of amino acids (AA). Even if some AA can be found on the surface of bioinspired inorganic NPs, it is difficult to apply these algorithms to these NPs.

So many antibiotics are derivatives of other antibiotics because new drug discovery can be slow and unpredictable. The “lock and key” mechanisms that dominate the interactions between biological molecules are used by drug scientists to create drugs that can fight bacteria and viruses in many preferred ways. Nanoparticles can be used to interrupt infections, but it was unclear how to move from the abstract concept of using nanoparticles to practical application. By applying mathematical methods to protein-protein interactions, it is now possible to streamline the design of nanoparticles. Unlike biomolecules, nanoparticles are more stable and can lead to new antibacterial and viral drugs.

The aim is to discover structural descriptors that can be used to universally describe complexes between proteins and NPs by examining the role of various structural aspects in the formation of protein-protein complexes. Identifying such descriptors would extend knowledge gained from huge PPI datasets and existing methods to NP-protein combinations observed in various biomedical scenarios, from drug delivery to environmental consequences of NPs.

The new ML method links nanoparticles to proteins using three alternative descriptions. The two structure-related descriptions were much more critical than the standard chemical description in predicting whether a nanoparticle will be an exact lock and crucial match to a particular protein. These two structural descriptions captured the complicated surface of the protein and how it might rearrange itself to facilitate key adjustments.

Chirality is a key consideration in predicting how proteins and nanoparticles will bind. Chirality is the clockwise or counter-clockwise rotation of a molecule or particle. Various proteins can be targeted, both outside and inside bacteria. These nanoparticles and proteins can be discovered using this model as a first screening step. Researchers can perform more in-depth simulations and tests based on this ML algorithm that has detected matches.

Machine learning algorithms like this will provide a design tool for nanoparticles, which can be used in various biological processes. A notable example is inhibiting the virus that causes COVID-19. The researchers believe that nanoparticles with broad antiviral activity can be designed effectively. Understanding and designing new classes of antibiotics with multiple modes of action is precisely the purpose of this work.



Sherry J. Basler