Novel chemometric approaches towards handling biospectroscopy datasets

Abstract

Chemometrics allows one to identify chemical patterns using spectrochemical information of biological materials, such as tissues and biofluids. This has fundamental importance to overcome limitations in traditional bioanalytical analysis, such as the need for laborious and extreme invasive procedures, high consumption of reagents, and expensive instrumentation. In biospectroscopy, a beam of light, usually in the infrared region, is projected onto the surface of a biological sample and, as a result, a chemical signature is generated containing the vibrational information of most of the molecules in that material. This can be performed in a single-spectra or hyperspectral imaging fashion, where a resultant spectrum is generated for each position (pixel) in the surface of a biological material segment, hence, allowing extraction of both spatial and spectrochemical information simultaneously. As an advantage, these methodologies are non-destructive, have a relatively low-cost, and require minimum sample preparation. However, in biospectroscopy, large datasets containing complex spectrochemical signatures are generated. These datasets are processed by computational tools in order to solve their signal complexity and then provide useful information that can be used for decision taking, such as the identification of clustering patterns distinguishing disease from healthy controls samples; differentiation of tumour grades; prediction of unknown samples categories; or identification of key molecular fragments (biomarkers) associated with the appearance of certain diseases, such as cancer. In this PhD thesis, new computational tools are developed in order to improve the processing of bio-spectrochemical data, providing better clinical outcomes for both spectral and hyperspectral datasets.