Thesis

Computer-aided drug design of bicyclic-derived BRD4 inhibitors

Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2023
Thesis identifier
  • T16764
Person Identifier (Local)
  • 201979893
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • The identification of new drug candidates is crucial due to high failure rates and prolonged research and development (R&D) timelines. Clinical attrition proves to be the main determinant in R&D efficiency; therefore, utilisation of appropriate enabling technologies should be implemented to address this major concern. Artificial intelligence (AI) has the capacity to revolutionise the drug discovery continuum by expediting hit identification and the overall optimization process. By taking a de novo drug design approach over traditional methods, it is possible to obtain a drug candidate with a balance of potency and physicochemical properties in a cost-efficient timeframe. This thesis focuses on the development of an AI-enabled drug discovery platform for the design and synthesis of 3,5-dimethylisoxazole analogues as bromodomain and extra-terminal (BET) protein inhibitors (Figure 1). Through collaborative efforts, a set of predictive quantitative structure-activity relationship (QSAR) models were developed from a diverse set of BRD4 inhibitors provided by GSK to be implemented into a computer-led workflow. To date, this approach has been utilised to successfully identify a potent fragment-like scaffold, 68, with moderate potency and promising ligand efficiency (LE). This fragment has been the focus of a computer aided drug design (CADD) and QSAR strategy to access compounds of enhanced potency and pharmacokinetic (PK) properties. This thesis also describes synthetic efforts towards the optimisation and validation of the predictive models for both potency (pIC50) and lipophilicity (LogD), to give us confidence for its future application within the project. Modifications to the molecular generator component of the platform were made to help reduce computational cost and time. By altering the computational workflow, more than 100 million molecules can be generated, with reasonable computational resource, then evaluated using the QSAR machine learning models to prioritise synthesis. A new series of compounds have been identified utilising this modified platform. This work remains in the early stages of development whereby re-introduction of an element of active or reinforcement learning will be required.
Advisor / supervisor
  • Jamieson, Craig
Resource Type
Note
  • This thesis was previously held under moratorium from 2nd November 2023 until 2nd November 2025.
DOI
Funder
Embargo Note
  • The digital copy of this thesis is restricted to Strathclyde users only until 2nd November 2028.

Relations

Items