Predicting coated-nanoparticle drug release systems with perturbation-theory machine learning (PTML) models

Area of Science:

• Nanotechnology
• Biomaterials Science
• Computational Chemistry
• Machine Learning

Background:

• Nanoparticle Drug Delivery Systems (DDNS) are crucial in nanotechnology and biomaterials.
• Existing experimental data for DDNS is dispersed and limited compared to component data.
• Current Machine Learning (ML) models often predict specific drug or nanoparticle activities, not integrated DDNS design.

Purpose of the Study:

• To develop a novel Perturbation Theory and Machine Learning (PTML) algorithm for designing new DDNS.
• To predict optimal combinations of nanoparticles (NP), coating agents, and drugs for DDNS.
• To create a multi-label model for selecting DDNS components with desired activity/toxicity profiles.

Main Methods:

• Downloaded and pre-processed large datasets of drug and coated Metal Oxide Nanoparticle (MONP) preclinical assays.
• Applied Perturbation Theory operators to account for structural, physicochemical, and assay condition variations.
• Integrated data using information fusion to create a comprehensive DDNS dataset (>500,000 pairs).

Main Results:

• Developed and trained a multi-label PTML model for DDNS component selection.
• Demonstrated the model's ability to predict optimal drug, coating agent, and NP combinations.
• Established a foundation for designing DDNS with tailored activity and toxicity.

Conclusions:

• The PTML algorithm represents a significant advancement in DDNS design.
• This approach facilitates the selection of components for optimized DDNS with desired therapeutic outcomes.
• The study provides the first multi-label PTML model for comprehensive DDNS component selection.