Artificial intelligence modeling and investigation of metal organic frameworks in drug delivery: modeling of loading capacity and toxicity behavior

TL;DR

This paper uses advanced AI models to predict how well metal organic frameworks can deliver drugs and their toxicity, showing highly accurate results.

Contribution

The study introduces Deep Gaussian Process Regression for modeling drug loading and toxicity in MOFs with high accuracy.

Findings

DGPR achieved R² scores of 0.99878 for drug loading capacity and 0.99911 for cell viability.

SHAP analysis provided insights into feature contributions for model predictions.

Cheetah Optimizer improved hyperparameter tuning for model performance.

Abstract

In this work, our aim was to develop predictive models of Drug Loading Capacity (g/g) and Cell Viability (%) in MOFs (Metal Organic Frameworks) for evaluation of these materials in drug delivery applications and assess their performance. We employed Gaussian Process Regression (GPR) and its advanced variants: Sparse Gaussian Process Regression (SGPR) and Deep Gaussian Process Regression (DGPR) as the base of our modeling framework to estimate the target values. The dataset was carefully preprocessed, involving outlier detection using the z-score method and normalization with Min-Max scaling approach. Dimensionality reduction was executed using Principal Component Analysis (PCA), while hyperparameter optimization was performed with the Cheetah Optimizer (CO), a metaheuristic method. Among the models evaluated, DGPR demonstrated superior performance, achieving mean cross-validation R
2 scores of 0.99878 ± 0.000092 for Drug Loading Capacity and 0.99911 ± 0.000127 for Cell Viability. Explainable AI techniques, especially SHAP, were employed to elucidate the model’s predictions, offering essential insights into the contributions of various features to the outcomes.

Figures