Tablet formulation design using response surface methodology and boosting trees
Yoshihiro Hayashi, Visiting Assistant Professor, Department of Pharmaceutical Design, Faculty of Medicine and Pharmaceutical Sciences (Pharmacy), Graduate School of Toyama University Yoshinori Onuki, Visiting Professor, Department of Pharmaceutical Design, Faculty of Medicine and Pharmaceutical Sciences (Pharmacy), Graduate School of Toyama University
Because the physical properties of the active pharmaceutical ingredient contained in a tablet have a significant impact on tablet quality, it is important to clarify the relationship between active pharmaceutical ingredient properties and tablet physical properties in tablet formulation design. However, understanding the properties of the active pharmaceutical ingredient requires evaluating the physical properties from multiple angles, and it is a complex system in which synergistic and antagonistic effects occur between the various physical properties. Therefore, the use of statistical analysis is useful for understanding the relationship between active pharmaceutical ingredient properties and tablet physical properties. In this presentation, we will introduce two research examples in which response surface methodology and boosted trees were applied to formulation design.
1. Evaluation of the correlation between response surfaces of drug substance combination tablets and placebo tablets. Several types of tablets with different excipient ratios were prepared according to an experimental design, and response surfaces for tablet hardness and disintegration time were created. Quantitative comparison of each response surface revealed a correlation between the response surfaces of drug substance combination tablets and placebo combination tablets. Applying these results and utilizing data from placebo combination tablets, we demonstrated the possibility of estimating the response surface of drug substance combination tablets from limited experimental points.
2. Evaluation of the relationship between drug substance properties and tablet physical properties using boosted trees
Various drug substance properties (molecular weight, partition coefficient, particle size distribution, bulk density, Hausner ratio, moisture content, elastic recovery, etc.) were evaluated for 81 types of drug substances. Then, by applying boosted trees, a machine learning function included in JMP Pro, the relationship between drug substance properties and tablet hardness was modeled with high accuracy. This study demonstrated that boosted trees are a useful method for clarifying important drug substance properties.
