Multi-Objective Optimization in Metabolomics/Computational Intelligence
Date
Authors
Advisors
Journal Title
Journal ISSN
ISSN
DOI
Volume Title
Publisher
Type
Peer reviewed
Abstract
The development of reliable computational models for detecting non-linear patterns encased in throughput datasets and characterizing them into phenotypic classes has been of particular interest and comprises dynamic studies in metabolomics and other disciplines that are encompassed within the omics science. Some of the clinical conditions that have been associated with these studies include metabotypes in cancer, in ammatory bowel disease (IBD), asthma, diabetes, traumatic brain injury (TBI), metabolic syndrome, and Parkinson's disease, just to mention a few. The traction in this domain is attributable to the advancements in the procedures involved in 1H NMR-linked datasets acquisition, which have fuelled the generation of a wide abundance of datasets. Throughput datasets generated by modern 1H NMR spectrometers are often characterized with features that are uninformative, redundant and inherently correlated. This renders it di cult for conventional multivariate analysis techniques to e ciently capture important signals and patterns. Therefore, the work covered in this research thesis provides novel alternative techniques to address the limitations of current analytical pipelines. This work delineates 13 variants of population-based nature inspired metaheuristic optimization algorithms which were further developed in this thesis as wrapper-based feature selection optimizers. The optimizers were then evaluated and benchmarked against each other through numerical experiments. Large-scale 1H NMR-linked datasets emerging from three disease studies were employed for the evaluations. The rst is a study in patients diagnosed with Malan syndrome; an autosomal dominant inherited disorder marked by a distinctive facial appearance, learning disabilities, and gigantism culminating in tall stature and macrocephaly, also referred to as cerebral gigantism. Another study involved Niemann-Pick Type C1 (NP-C1), a rare progressive neurodegenerative condition marked by intracellular accrual of cholesterol and complex lipids including sphingolipids and phospholipids in the endosomal/lysosomal system. The third study involved sore throat investigation in human (also known as `pharyngitis'); an acute infection of the upper respiratory tract that a ects the respiratory mucosa of the throat. In all three cases, samples from pathologically-con rmed cohorts with corresponding controls were acquired, and metabolomics investigations were performed using 1H NMR technique. Thereafter, computational optimizations were conducted on all three high-dimensional datasets that were generated from the disease studies outlined, so that key biomarkers and most e cient optimizers were identi ed in each study. The clinical and biochemical signi cance of the results arising from this work were discussed and highlighted.