Thesis

Application of deep generative models for the design of pharmaceutical manufacturing processes

Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2025
Thesis identifier
  • T17214
Person Identifier (Local)
  • 202155680
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Designing processes for pharmaceutical product manufacturing is a complex andresource-intensive task. With increasing research costs and quality standards, thepharmaceutical industry seeks innovative technologies to enhance productivity andmaintain competitiveness. While a variety of tools exist in the process design domain foroptimizing conditions or selecting materials, options for guiding the selection ofmanufacturing operations remain limited.In this context, deep generative models (DGMs) emerge as a promising approach.DGMs, known for learning the probability distribution of data, have gained popularity fortheir ability to generate realistic examples, commonly applied in text and imagegeneration. In drug discovery, DGMs have successfully generated new active substanceswith desirable properties. However, their application in the manufacturing space remainsunexplored. These models have the potential to assist in operation selection andexperimental targeting, thereby reducing development time.This thesis aims to investigate the applicability of DGMs in pharmaceutical manufacturingprocess design, developing DGMs capable of generating plausible sequences ofoperations for product manufacturing, taking input information about the target product.A significant challenge in developing DGMs is the requirement for large datasets. Toaddress this, two datasets were constructed using natural language processing (NLP)applied to primary and secondary manufacturing data extracted from patents. Theprimary processing dataset comprises over 385K manufacturing processes, while thesecondary processing dataset includes approximately 9K procedures for various dosageforms and active ingredients.The study involved training and comparing several architectures based on generativeadversarial networks (GAN) and variational autoencoder (VAE) using different metrics.Real and generated sequences were contrasted manually to evaluate how closely themodel outputs resemble typical manufacturing sequences. This research contributes tothe exploration of DGMs’ application in pharmaceutical manufacturing, offering insightsinto their potential for operation selection and product development. In the end, DGMswere successfully trained and their potential for the generation of plausible sequenceswas demonstrated. A survey assessed by a panel of experts yielded that the models generated sequences at least as good as the actual procedures in 38% of occasions forthe primary domain. While this shows the potential of generative modelling in this field, italso remarks there is room for improvement to make it applicable in real-world scenarios.
Advisor / supervisor
  • Brown, Cameron
  • Johnston, Blair
Resource Type
DOI
Date Created
  • 2024

Relations

Items