Computational pharmaceutics approaches to inform drug developability: focus on lipid-based formulations

Loading...
Thumbnail Image
Files
Bennett-LenaneH_Partial.pdf(4.57 MB)
Partial Restriction
Date
2021-10-04
Authors
Bennett-Lenane, Harriet
Journal Title
Journal ISSN
Volume Title
Publisher
University College Cork
Published Version
Research Projects
Organizational Units
Journal Issue
Abstract
Purpose: Declining productivity in the face of increasing numbers of poorly water-soluble drugs has fast-tracked necessity for predictive tools which assess the delivery potential of bio-enabling formulations. However, there is a perceived risk associated with early-stage selection of bio-enabling formulations. Computational pharmaceutics is a growing area of research interest to support structured guidance in formulation strategy. Using data-driven modelling, a streamlined roadmap of computational possibilities for development scientists is possible. Accordingly, the aim of this thesis was to examine the application of machine learning (ML) computational modelling to inform candidate developability. Via prediction of both quality target product profile characteristics and formulation performance indicators for lipid-based formulations (LBF). In recognition of the fact that computational models will not entirely circumvent need for manual screening, a further aim was to explore if analysis of landrace pig gastrointestinal fluids could facilitate increased bio-predictive performance of in vitro tools for LBFs. Methods: Data-driven computational models using various ML algorithms, with both classification and regression outputs, were developed to predict food effect on bioavailability, solubility ratio (SR) upon self-emulsifying drug delivery system (SEDDS) dispersion and apparent degree of supersaturation (aDS) ratio in supersaturated LBFs (sLBF). Model performance was validated using test sets or comparisons to ex vivo results. Gastrointestinal fluids from the landrace pig were collected in the fasted state, fed state and post placebo SEDDS administration. Ex vivo solubility analysis and microscopic imaging were completed using these fluids, where in vitro biorelevant dispersion screening with SEDDS using various dilution conditions was compared to the ex vivo results. Results: Firstly, this thesis demonstrated the applicability of ML for the prediction of a quality target product profile characteristic of interest in early development, namely food effect on bioavailability. Secondly, this thesis has advanced computational pharmaceutics to inform drug developability. Computational predictions of solubility gain upon SEDDS dispersion informed a biopharmaceutical dose number in intestinal fluids, which can be incorporated within the developability classification system (DCS) framework to inform drug developability. Thirdly, in recognition of the use of supersaturated LBFs to overcome dose loading limitations, this thesis has demonstrated how ML algorithms can predict the maximum dose loading upon thermal induced supersaturation. Moreover, increased understanding of the fate of SEDDS upon oral administration furthered the utility of the pig pre-clinical model, validated accuracy of in silico predictions and aided development of a bio-predictive in vitro screening tool for LBFs. Ultimately, it was demonstrated that the computational and in vitro tools developed in this thesis can be embedded within a wider refined drug substance to drug product development framework. Conclusion: This thesis highlighted how pharmaceutics datasets are amenable to ML. The ability of computational pharmaceutics to facilitate structured formulation decisions was demonstrated. As model development aided increased understanding of the investigated phenomena through their relationship to drug properties, this thesis identified the significant potential to be gained from early analysis of drug properties. Additionally, utility of the landrace pig model to inform increasingly bio-predictive in vitro screening tools was established. The proposed refining of the drug substance to drug product development framework demonstrated the significance of both computationally informed and experimentally confirmed aspects of drug developability decision-making.
Description
Keywords
Machine learning , Lipid-based formulations , Drug development , Computational modelling , Poorly-water soluble drugs
Citation
Bennett-Lenane, H. 2021. Computational pharmaceutics approaches to inform drug developability: focus on lipid-based formulations. PhD Thesis, University College Cork.