Prediction of Drug Dissolution from Sustained-Release Pellet by a Portable Near-Infrared Spectrometer

Sijun Wu,Yinglian Yang,Li Wang,Chaoliang Jia,Zhiqi Guan,Haiyang Chen,Yadong Zhu,Wenlong Li

DOI: https://doi.org/10.1016/j.jddst.2024.106424

IF: 5

2024-01-01

Journal of Drug Delivery Science and Technology

Abstract:Dissolution properties are critical assessment targets to be examined in the development of sustained-release pellets. Comprehensively understanding the dissolution behavior of the drug, followed by appropriate formulation adjustment, is essential to enhance the product quality consistency. In order to achieve the rapid prediction of the dissolution behavior of the sinomenine hydrochloride (SH) sustained-release pellets, a method based on the portable near-infrared (NIR) spectrometer combined with machine learning was proposed. Seventy-eight batches of pellets with different dissolution behaviors were designed for the research. Principal component analysis, isolation forest, and Mahalanobis distance were used for the outlier detection. In this study, the dissolution profile prediction models were developed using partial least squares regression (PLSR), Gaussian process regression (GPR), and convolutional neural network (CNN) algorithms. Additionally, the impact of variable selection on model performance was investigated using competitive adaptive weighted resampling and grey wolf optimizer. Compared to the PLSR and GPR models, the grey wolf optimizer (GWO)-CNN model gave the best prediction of the dissolution profile with root mean square error of prediction values of the cumulative SH release for eight time points except the 5th being below 5.5% and RP values exceeding 0.95. Additionally, all validation samples and independent test samples exhibited f1 values below 15 and f2 values exceeding 50. The results demonstrated that the portable NIR spectrometer combined with machine learning was a promising tool used for the dissolution profile prediction of the SH sustained-release pellets.