DEVELOPMENT AND INVESTIGATION OF CHEMOMETRIC BASELINE CORRECTION APPROACHES AND METABONOMIC CLASSIFICATION ALGORITHMS

Abstract

Metabonomic analysis has been used for classification in a diverse range of areas from toxicology and dietary effects through to parasitology and molecular epidemiology, including disease diagnosis and therapy monitoring. Metabonomic data requires correction via pre-processing approaches followed by post-processing involving a robust modelling approach to provide accurate and fast classification. In this work, we developed novel algorithms for both phases. For pre-processing, we developed a baseline correction algorithm, Automated Iterative Moving Averaging (AIMA) , which has similar accuracy as existing semi-automated algorithms but is fully automated and computationally more efficient (28.6 to 197.7 times faster). For post-processing, we developed a fully automated classification algorithm, Automated Pearson?s correlation change classification (APC3) , which has similar or better prediction accuracy as the current state of art algorithms for metabonomic data but is 3.9 to 7 times faster. Finally, we did a comparative study on four sparsity embedded classification techniques.