Modelling approaches for rational solvent selection in drug development, enhancing the solubility prediction of small molecules

Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2019
Thesis identifier
  • T15704
Person Identifier (Local)
  • 201456254
Qualification Level
Qualification Name
Department, School or Faculty
  • In the pharmaceutical industry crystallisation is the preferred method used to control crystal size and particle growth, and to purify the final product. It can also be used for the isolation of any impurities. Using the minimum amount of material is desirable. Therefore, the ability to predict solubility and to construct accurate and robust models for complex molecules has been of ongoing interest in the pharmaceutical industry. There are several methods to predict solubility in silico, including: COSMOtherm, NRTL-SAC, UNIFAC, and SAFT-γ Mie.;This work will focus on the use and appraisal of ab initio method COSMOtherm and the application of a "correction factor" using the machine learning algorithm random forest to improve accuracy of predictions. Chapter Two compares experimental data with COSMOtherm to assess the robustness and reliability of the method. The influence of adjustable parameters required for predictions: enthalpy of fusion, and melting temperature, were assessed. These studies detail the importance of accurate measurements of these parameters and how deviations from their true value can affect the accuracy of solubility predictions.;Chapter Three details the building of a linear regression model using a design of experiment approach for almost instantaneous predictions using no specialised software, for non-experts and modellers. Chapter Four details the building of machine learning models using random forest to apply a correction factor to the error between COSMOtherm and experimental data. Chapter Five uses predictive methods and a workflow approach to select crystallisation and wash solvents. A case study using paracetamol and its impuritiesis considered.
Advisor / supervisor
  • Johnston, Blair
  • Florence, Alastair
Resource Type
Date Created
  • 2019
Former identifier
  • 9912922790202996