Date of Award

2017

Degree Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering

Department

Electrical, Computer, and Biomedical Engineering

First Advisor

Jay Wang

Abstract

Biological processes are complex and can be modeled using combination of linear and non-linear models. During the cell culture process, cells may change or evolve and it is extremely important to understand the variability of the process to manufacture consistent product. In order to maximize the output from the process, process parameters need to be characterized and optimized. Data from process involves linear and non-linear patterns and some of the parameters are auto-correlated. One of the objectives of study was to compare Unsupervised Dimesional Reduction methods to Supervised machine learning algorithms applied to biopharmaceutical manufacturing process data and suggest a new 2-stage approach including a combination of unsupervised and supervised algorithms for better predictability. Analytical methods are used to measure the quality of the product. Main objective of methods transfer is to avoid release of product that does not meet specifications as well as avoid rejection of good product. The effect of sample size for establishing analytical method equivalency and comparison of statistical methods during assay transfers was performed and criteria for out of specification risk mitigation was recommended. Shelf life of a biopharmaceutical product is typically based upon the stability data. Factors that impact stability of a product and shelf life were studied in detail using multiple statistical models and criteria for choosing the appropriate model was recommended.

Share

COinS