(148a) Deployment of Machine Learning Models in Pharmaceutical Development

The advantage of model driven decisions in research organizations has long been recognized and applied in the pharmaceutical research and development. From the conceptualization of the Quality by Design framework as a series of mathematical constructs [1,2] to the application of digital twins [3]. Machine Learning methodologies are widely applicable and efficient tools to construct mathematical models of relevant pharmaceutical processes enabling process optimization, robustness characterization, and failure rate estimation. However, the widespread adoption of these approaches requires a combination of streamlined availability, and institutional know-how to interpret and incorporate the results in the decision-making process.

This talk addresses different approaches to leverage emerging and stablished tools in Machine Learning to pharmaceutical development applications by comparing the adoption of specific algorithms and computational platforms. Specifically, a generic workflow for data exploration analysis, design of experiments, feature selection, and model exploration has been developed and deployed with a combination of cloud computing and Jupyter notebooks. In our experience this approach has been more successful at integrating a model-based decision making culture in process development than previously attempted alternatives.

[1] Peterson, J.J., 2008. A Bayesian approach to the ICH Q8 definition of design space. Journal of biopharmaceutical statistics, 18(5), pp.959-975.

[2] Garcia-Munoz, S., Luciani, C.V., Vaidyaraman, S. and Seibert, K.D., 2015. Definition of design spaces using mechanistic models and geometric projections of probability maps. Organic Process Research & Development, 19(8), pp.1012-1023

[3] Chen, Y., Yang, O., Sampat, C., Bhalode, P., Ramachandran, R. and Ierapetritou, M., 2020. Digital Twins in Pharmaceutical and Biopharmaceutical Manufacturing: A Literature Review. Processes, 8(9), p.1088.