Sequential EXtreme Gradient Boosting-Based Descriptor Reduction for Size Prediction of Zwitterionic Polymer-Based...

Area of Science:

• Materials Science
• Computational Chemistry
• Drug Delivery

Background:

• Zwitterionic polymers (ZP) are crucial in drug delivery systems (DDSs).
• Controlling the size of ZP-based DDSs is vital for efficacy.
• Understanding structural influences on ZP self-assembly and size is needed.

Purpose of the Study:

• To investigate how structural descriptors impact the size of zwitterionic polymer-based drug delivery systems.
• To develop and validate a machine learning approach for predicting nanoparticle size.
• To identify key structural features governing ZP self-assembly and size.

Main Methods:

• A novel descriptor reduction strategy, Sequential XGBoost (SXGB), was developed to streamline 312 molecular descriptors to 11 key features.
• Machine learning models, including SXGB, were trained and tested on a curated dataset.
• Local Interpretable Model-agnostic Explanations (LIME) were used for descriptor interpretability, and data augmentation enhanced model robustness.

Main Results:

• The SXGB model achieved high accuracy in predicting nanoparticle size, with R² values of 84.2% (training) and 80.9% (testing).
• pH was identified as the most influential descriptor affecting zwitterionic polymer nanoparticle size.
• The SXGB model outperformed supporting vector regression and random forest models in predictive accuracy.

Conclusions:

• The SXGB machine learning approach effectively predicts zwitterionic polymer nanoparticle size for drug delivery applications.
• Structural descriptors, particularly pH, significantly influence nanoparticle self-assembly and final size.
• This study provides a robust computational framework for designing optimized ZP-based drug delivery systems.