Prediction of Anti-Glioblastoma Drug-Decorated Nanoparticle Delivery Systems Using Molecular Descriptors and Machine...

Area of Science:

• Computational chemistry
• Nanomedicine
• Machine learning in drug discovery

Background:

• Predicting drug-decorated nanoparticles (DDNPs) is crucial for medical applications.
• Developing effective glioblastoma treatments remains a significant challenge.

Purpose of the Study:

• To build and evaluate Perturbation Theory Machine Learning (PTML) models for predicting DDNPs with anti-glioblastoma activity.
• To identify promising drug-nanoparticle combinations for glioblastoma therapy through virtual screening.

Main Methods:

• Utilized molecular descriptor perturbations of drugs and nanoparticles as inputs for PTML models.
• Integrated experimental nanoparticle data with ChEMBL drug assay data.
• Tested ten machine learning algorithms, selecting 41 features for 855,129 drug-nanoparticle complexes.

Main Results:

• The best performing model was a Bagging classifier ensemble of 20 decision trees.
• Achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.96.
• Attained 87% accuracy on the test subset for predicting anti-glioblastoma activity.

Conclusions:

• The developed PTML model demonstrates high predictive power for anti-glioblastoma DDNPs.
• This approach facilitates efficient virtual screening of potential glioblastoma drug-nanoparticle therapies.
• The study provides reproducible datasets and scripts for further research.