Implementing partial least squares and machine learning regressive models for prediction of drug release in targeted drug delivery application

Abstract

A combined methodology was performed based on chemometrics and machine learning regressive models in estimation of polysaccharide-coated colonic drug delivery. The release of medication was measured using Raman spectroscopy and the data was used for estimation of drug delivery using machine learning models. Raman data was used along with some inputs including coating type, medium, and release time to estimate the drug release as the sole target. The study explores predictive modeling for the release variable in a dataset including 155 samples with over 1500 spectral variables. Partial least squares (PLS) was applied for dimensionality reduction, and models such as AdaBoost with linear regression, multilayer perceptron (MLP), and Theil-Sen regression were utilized, achieving the highest predictive performance with the AdaBoost-MLP model (R2 = 0.994, MSE = 0.000368). Uniquely, this work integrates glowworm swarm optimization (GSO) for model hyperparameter tuning, enhancing model accuracy and efficiency. The results suggest that spectral characteristics combined with environmental and compositional factors provide a comprehensive foundation for modeling release dynamics in evaluation of targeted colonic delivery formulations.