DM2 platform II : AI-assisted optimization of oral solid dosage form development

Salehian, M and Moores, J and Abbas, F and Markl, D (2022) DM2 platform II : AI-assisted optimization of oral solid dosage form development. In: CMAC Annual Open Day 2022, 2022-05-16 - 2022-05-18.

[thumbnail of Salehian_etal_CMAC_2022_DM2_platform_II_AI_assisted_optimization_of_oral_solid_dosage]
Preview
 Text. Filename: Salehian_etal_CMAC_2022_DM2_platform_II_AI_assisted_optimization_of_oral_solid_dosage.pdf
Final Published Version
License: Strathprints license 1.0
Download (790kB)| Preview

Abstract

Digital Medicines Manufacturing (DM2) platform II aims to leverage AI to develop an autonomous workflow for drug product manufacturing and testing system by identifying optimal formulation and process parameters that deliver desired critical quality attributes (CQAs). This approach aims to de-risk and accelerate drug product development by reducing experiments, development time, and materials use by 60%. The initial objective is to develop a database of hundreds of historical data and new experiments. This database is then used to develop a hybrid machine for predicting product (tablets and capsules) attributes by utilizing domain knowledge (empirical/mechanistic models) and AI-powered models (where domain knowledge is not available/reliable). The hybrid machine is then integrated into an iterative, model-informed optimization framework, to smartly plan experiments that drive the automated manufacturing and testing system, which will be used to collect multi-scale and –point data and update the model(s) to learn from the experiments. Fit-for-purpose optimization algorithms will be employed to execute this loop based on the objective of experiments (i.e. exploration of the knowledge space or exploitation of the model-based optimization) and it continues until the targets are achieved. The proposed smart experimental planning method has been tested using the Gurnham (empirical) model as a predictor of porosity based on peak compression pressure. The initial analysis of the uncertainty of fit demonstrates that adding only 1 data point results in a 20-fold improvement in the accuracy of prediction while adding 8 more data points leads to a minimal improvement, highlighting the significance of the here-developed smart experimental planning procedure in achieving acceptable prediction accuracy at a minimal experimentation cost.

Persistent Identifier

https://doi.org/10.17868/strath.00081198
CORE (COnnecting REpositories)