AI-Based prediction and monitoring of drug formulation research and development

Abstract

When new solid formulations reach a stage of research and development (R&D), the number of experiments to be performed is generally several times higher than expected such that the R&D cost will surge. It is noteworthy that the quality evaluation standard for solid formulations and the dissolution test results take on critical significance to guiding the solid formulation research, whereas the experimental phenomenon of dissolution is difficult to be fully observed by experimenters for the extremely long experiment time of at least 8 h. Thus, a system exhibiting prediction and phenomenon monitoring functions is developed and tested in this study to tackle down the above-described problems. First, a computer technology-based prediction system for the dynamic solubility of active pharmaceutical ingredient (API) is developed in accordance with the solid formulation R&D process, where a novel equation is employed, instead of polynomial regression. After the experimentally acquired dynamic solubility data are trained, the developed system is capable of predicting the dynamic solubility rate of API following the blade dissolution speed. Moreover, a novel tablet dissolution rate prediction system is built using artificial neural networks (ANN) and non-linear regression methods. Compare with SVM, ANN can offer a faster prediction speed and suitable for the future big data model. This system can predict dissolution rates with ANN and processing the prediction results based on two novel non-linear regression methods. Accordingly, the demanded system database becomes less than the orthogonal design. Compared with previously developed systems, the built system exhibits higher prediction precision with a less amount of training data. Besides, the system, based on the novel input data design, is enabled to use the data whose formulation composition resembles prediction composition. Furthermore, the input data screening function is introduced into the system based on the prediction function to avert the effect of experimental error. Lastly, a phenomenon monitoring system covering a camera module and an image recognition module is developed to monitor the dissolution phenomenon of tablets during the dissolution test. The image recognition program based on the region growth, Hue, Saturation, and Value (HSV) is capable of capturing and pixelating tablets of different colors during their movement in the dissolution cup. To be specific, the camera module comprises a visible/infrared light camera and an infrared light source, such that the monitoring system is enabled to recognize tablets in dark and bright environments. Using ANN, non-linear regression, and image recognition methods, the dynamic solubility of API is predicted, the tablets dissolution rate is predicted based on a smaller database, and the real-time dissolution phenomenon is monitored in a wide variety of environments in this study.

Date of Award	Jul 2023
Original language	English
Awarding Institution	University of Nottingham
Supervisor	C.F. Kwong (Supervisor), David Cho Siu-Yeung (Supervisor), Boon Giin Lee (Supervisor) & Liang Huang (Supervisor)