| Popis: |
Mass Spectrometry Imaging (MSI) is a sensitive analytical tool for detecting and spatially localising thousands of ions generated across intact tissue samples. The datasets produced by MSI are large both in the number of measurements collected and the total data volume, which effectively prohibits manual analysis and interpretation. However, these datasets can provide insights into tissue composition and variation, and can help identify markers of health and disease, so the development of computational methods are required to aid their interpretation. To address the challenges of high dimensional data, randomised methods were explored for making data analysis tractable and were found to provide a powerful set of tools for applying automated analysis to MSI datasets. Random projections provided over 90% dimensionality reduction of MALDI MSI datasets, making them amenable to visualisation by image segmentation. Randomised basis construction was investigated for dimensionality reduction and data compression. Automated data analysis was developed that could be applied data compressed to 1% of its original size, including segmentation and factorisation, providing a direct route to the analysis and interpretation of MSI datasets. Evaluation of these methods alongside established dimensionality reduction pipelines on simulated and real-world datasets showed they could reproducibly extract the chemo-spatial patterns present. |
