Filters
Availability
Publication Year
Resource Type
Faculty
School/Institute
Publication Search Results
Sort
Results per page
1 - 3 of 3
-
(2022)ThesisAdvances in molecular biology data collection, leading to the accumulation of large amounts of diverse data, call for novel computational approaches to enable their effective analysis. This thesis explored the application of visual-analytics-driven bioinformatics approaches to four biomolecular data-driven challenges. For analysing time-series omic and multiomic data, a novel method, Minardo-Model, was developed. Minardo-Model can identify key events (e.g. phosphorylation) from such time-series data and temporally order them. To visualise the inferred order of events, two novel visualisation approaches, event maps and event sparklines, were developed. Minardo-Model was tested using two time-series datasets and in both cases, the event orderings derived by this method correlated with prior knowledge. To streamline the use of experimental 3D protein structures for analysing sequence variants, a novel method was developed and integrated into Aquaria. For variants specified in the HGVS notation, the method identifies and displays a best matching structure. Additionally, for each variant specified, all structures spanning the variant, and containing the exact variant (missense only), along with sequence features retrieved from external resources, are summarised. The developed approach was used to analyse variants in human ACE2, and SARS-CoV-2 spike, revealing novel insights. For pathogenic bacterial isolates characterised using multilevel genome typing (MGT), the MGTdb web service was developed. MGTdb, enables upload of isolates as sequence reads or extracted alleles, which are processed and assigned the MGT-identifiers. The features of MGTdb, such as interactive visualisation tools, data download and export to external software, enable epidemiological exploration in the context of the local or global database of isolates. The usability of MGTdb was successfully demonstrated through three case studies. For identifying insertion sequences (IS) from short-read sequencing data, a novel method, WiIS, was developed. WiIS was tested on Bordetella pertussis isolates, for which both short-read (test data) and long-read sequences (ground truth) were available - WiIS was found to have high precision and recall. It also outperformed other published tools in identifying IS in B. pertussis genomes. The novel bioinformatics methods developed in this thesis enable novel analysis of a wide variety of data thus providing insight into various biomolecular processes.
-
(2023)ThesisImproving the efficiency, reliability, and durability of photovoltaic cells and modules is key to accelerating the transition towards a carbon-free society. With tens of millions of solar cells manufactured every day, this thesis aims to leverage the available characterisation data to identify defects in solar cells using powerful machine learning techniques. Firstly, it explores temperature and injection dependent lifetime data to characterise bulk defects in silicon solar cells. Machine learning algorithms were trained to model the recombination statistics’ inverse function and predict the defect parameters. The proposed image representation of lifetime data and access to powerful deep learning techniques surpasses traditional defect parameter extraction techniques and enables the extraction of temperature dependent defect parameters. Secondly, it makes use of end-of-line current-voltage measurements and luminescence images to demonstrate how luminescence imaging can satisfy the needs of end-of-line binning. By introducing a deep learning framework, the cell efficiency is correlated to the luminescence image and shows that a luminescence-based binning does not impact the mismatch losses of the fabricated modules while having a greater capability of detecting defects in solar cells. The framework is shown in multiple transfer learning and fine-tuning applications such as half-cut and shingled cells. The method is then extended for automated efficiency-loss analysis, where a new deep learning framework identifies the defective regions in the luminescence image and their impact on the overall cell efficiency. Finally, it presents a machine learning algorithm to model the relationship between input process parameters and output efficiency to identify the recipe for achieving the highest solar cell efficiency with the help of a genetic algorithm optimiser. The development of machine learning-powered characterisation truly unlocks new insight and brings the photovoltaic industry to the next level, making the most of the available data to accelerate the rate of improvement of solar cell and module efficiency while identifying the potential defects impacting their reliability and durability.
-
(2021)ThesisMultiferroic BiFe0.5Cr0.5O3 (BFCO) in which ferroelectric and magnetic orders coexist has gained research interest owing to its potential applications, e.g., spintronic and resistive random-access memory. Moreover, multiferroics possess a narrower bandgap compared to typical ferroelectrics, extending their application to photovoltaic devices. In contrast to the conventional semiconductors, the polarization-induced electric field facilitates the photoexcited charge separation, leading to an above-bandgap photovoltage in ferroelectrics. Nevertheless, a long-standing issue is the relatively low absorption of visible light. Thus, it is essential but challenging to tune their bandgap without compromising ferroelectricity. This thesis explores structural phase transition in the epitaxial BFCO films grown on SrRuO3 buffered (001) SrTiO3 substrate via Laser Molecular Beam Epitaxy (LMBE). Reciprocal space mapping result shows strain relaxation mechanism is not solely by the formation of misfit dislocation but also by changing the crystal symmetry, transitioning from tetragonal-like to a monoclinically distorted phase as the thickness increases. The crystallographic evolution is also coupled with bandgap modulation, confirming that BFCO structure and its physical properties are strongly intertwined. Using spectroscopic ellipsometry, the slight redshift of the bandgap distinguishes the absorption process of the T-like BFCO layer from that of monoclinically distorted structure, further confirmed by spectral photocurrent measurement via conductive-atomic force microscopy. The preparation of pure phase BFCO film with a robust polarization is of paramount importance for practical application. Yet, similar to the parental bismuth ferrite, BFCO suffers from poor electrical leakage performance. We report a three-order of magnitude suppression in the leakage current for the BFCO film through judicious adjustment of the growth rate. Scanning probe microscopy (PFM, AFM and c-AFM) results reveal that both microstructure and ferroelectric properties can be tuned by lowering the growth rate, ensuing realization of the room-temperature ferroelectric polarization comparable to the ab-initio predicted value. This thesis provides a facile strategy to tailor the structure-property of epitaxial BFCO film and its functional response for emerging optoelectronic devices.