Data Driven Prediction of Dissolution Profiles and Optimization of Metformin Hydrochloride Sustained Release Tablets Using Deep Neural Network

Abstract

The optimization of Metformin HCl sustained-release (SR) tablet newlineformulation requires a data-driven approach combining Artificial Intelligence (AI) and newlineDesign of Experiments (DoE). This study begins with a collection of raw material data newlinefrom the year 2010-2024 to establish formulation parameters. AI models are then newlineselected using the TPOT AutoML approach from which six models were chosen such newlineas Decision tree regressor, Gradient Boost regressor, Extra Gradient Boost Regressor, newlineRandom Forest regressor, Extra tree regressor and Deep neural network which are newlinetrained on the data set for the feature selection. The best-predictive model was newlineidentified as Deep neural network from the six ML models with 10-fold crossvalidation. newlineFrom these models, key independent variables (polymer, binder, filler, newlineglidant, and lubricants) are selected to design formulations using the Box-Behnken newlineDoE. newlineThe suitable combinations obtained from DoE are further analyzed using newlinea Deep Neural Network (DNN) model to predict dependent variables, such as tablet newlinethickness, hardness, friability, drug content and in-vitro dissolution release. Based on newlinethese predictions, Metformin HCl SR tablets are formulated experimentally and newlinesubjected to physical and chemical evaluation to assess their compliance with newlinepredefined quality parameters newline