Data Platform and Modelling Drug Substance Cyber-Physical System

Crystalline solid-state structures are fundamental to understanding material properties such as solubility, stability, and polymorphism. However, the modelling and simulation of the solid state is typically computationally intensive and does not naturally lend itself to use in cyber-physical systems, which need to be able to use models and data in real time to operate effectively.

This PhD project aims to accelerate model applications, such as crystal structure prediction (CSP), by developing novel, information-rich fingerprints that effectively represent crystalline solid-state structures. These fingerprints will enable rapid similarity calculations using vector databases, significantly accelerating CSP processes.

The first objective is to design fingerprints that capture intricate structural details beyond conventional molecular fingerprints. Methods will be developed to optimise these fingerprints for predictive performance and computational speed. Benchmarking against existing molecular fingerprints will assess improvements in efficiency and accuracy.

The second objective focuses on leveraging these novel fingerprints to enhance data-driven predictions of molecular physical properties, particularly solubility. The project will investigate the capability of the fingerprints to handle crystal polymorphism and account for solubility variations arising from different solid forms. This involves rigorous testing against experimental data to validate predictive models.

Finally, the project will develop a comprehensive data model and ontology for the solid state. This ontology will encapsulate structural concepts and features, facilitating advanced natural language processing (NLP) searches for crystal structures with specific functionalities and geometries. Such a framework will improve data accessibility and support researchers in identifying materials with desired properties.

Expected Outcomes: - Creation of optimised, information-rich fingerprints for crystalline structures. - Validated methods for rapid similarity calculations to support CSP. - Enhanced predictive models for molecular properties like solubility, accommodating polymorphism effects. - A robust solid-state ontology to improve data retrieval and NLP search capabilities.

This research will significantly contribute to computational chemistry and materials science by providing rapid, innovative tools for structure prediction and property analysis, ultimately advancing the development of new materials and pharmaceuticals.